It was a work heavy and cozy week, had a long weekend. AI is here, going nowhere, so trying to learn it as much as I can. Its a bit of a tricky phase of world, nobody is sure what is the right way and what is a tumbling down hill. It is usually is like that in life, but this is extreme, people are questioning their existence and stating that AI is conscious, as we even don’t know what conscious means.

Quote of the Week

“The quieter you become, the more you are able to hear.”
— Rumi

This is such a lovely quote. I try to keep quite as much as possible, I am shy that is a different point. But I feel there is no need to speak in most of the situations, it just feels waste or just sugar coating. Maybe I am a bit blunt, I speak only when I feel something is wrong or the situation is out of hand.

But when we are quite, not just in the literal sense, from the mind, the thoughts are quieter, we can observe more carefully and open-mindedly. It is a reason why I can see and feel things many can’t. To be able to hear someone or something, you need to calm your self and the mind first, assumptions kills more things than anything else. The bias and the pre-concieved knowledge hurts more than bad knowledge.

Read

Surprisingly I didn’t read much this week. I spent more time working with the clankers. It was a bit unsettling at first, but then it becomes a habit to delegate everything to it. That doesn’t mean I didn’t read at all, I just didn’t read tech-related stuff. I read 2 book on the long weekend, all cozy and korean-japanese style lofi books. Because I just need a bit of time to escape this technical jungle. I don’t hate being there, but sometimes a fresh air is required to sustain. I read “Every day I read” and “The housekeeper and the Professor”, beautiful books.

1. Vibe coding and agentic engineering are getting closer more than I had like

   • This is true. The agents are producing so much code that a human cannot even keep up to. But that shouldn’t be a excuse to not learn what the changes have been made. At this point its about slowing down and letting ourselves savor what LLMs in the past 3 years have made us.

   • The other part is very interesting of not getting concerned about being replaced as a developer. The analogy of plumber hits right.

   • If one can quickly plumb his house things, why even hire a plumber. But we know that we don’t like doing that so we hire him eventually. But can this be true for software too, it has been the most sort of thing that bugs people. Coding is not trivial, yes writing code has become due to AI, but code was not the bottleneck. The patience level of a developer, the ability to reason through, to find the gaps, to have a taste, to understand human sentiment, it is what truly makes a true software developer if not less. And that AI would never be able to do (atleast for the next year maybe)

2. We programmed a program to program new programs (2011)

   • Imagine being 15 years ahead of time

   • Just banger. How can one have written this is 2011, what a pscyhosis moment for him/her it would have been.

Watched

• Andrej Karapathy: From Vibe Coding to Agentic Engineering

  • That was a good interview. His examples are great. The Menu Gen. Maybe its peak AI.

  • The analogy of custom UI was just min blowing, we wont fret over a bug, we will be fighting with versions now! Ew I don’t like where this is going.

• What’s Next for Software Tracking and Git?

  • Oh, codeberg, forgejo. That sounds familiar, I thought we might not need them. Oh! GitHub’s normal always available has just made me delusional I think. Yes that hurts, that we can view a person’s history over decades. Pufff! Gone.

  • The hiring for new software developers looks a little worisome. What will people look at? I know GitHub profile is not a metric, but still it showed the person’s familiarity with dealing with software.

• DHH’s new way of writing code

  • This is a bit off putting. I thought he might say, AI is joke, its a fad. Sigh!

  • The point that, I felt like engineering manager I though I need to manage people, but these AI Agents are different. I agree. If it would have been something like a real managerial work, I might have died the moment I had to do that. Some things are just not someone’s piece of cake.

  • Yes AI Agents are really addictive. But yes health and family comes first still.

  • The teams moving ahead might be leaner, there is no doubt, but the problem would be still the same, solve actual problems. There will be a explosion of business if that is the case, because it has to balance it out. More people in a company, less companies, less people in a company, more companies? I think this logic might work in the next 5 years?

• Why AI Agents are tither the best or worst thing we’ve built

  • That is wild. Absolute carnage.

  • AI Agents are a good thing, but as anything, it depends on how people use it. Like internet, like social media and intelligence. If set in people’s hand this can do crazy things.

  • The open claw moment is really the point for people to show the vastness and the craziness of what uncontrolled AI can do, the lethal trifiecta, Simon Wilison trademark.

• Am I Crazy? Wading Through AI Episode 3

  • Yes, I am crazy. These is a hype. It won’t make you a 10x, 100x developer. If that is the case then a person can do a year’s worth of work in a week. That is impossible. Because if that is the case, then send him on a vaccation for the rest of the year, have we seen that? Sine the work is done, he can rest in peace. But no! They give excuse of more work.

  • This is really a bit of hype. It will make you productive sure. But not the numerical one. It will help you massively in certain tasks, but slow you down terribly in others. Right task and the right tool still holds.

• Volatile keyword in Java

  • I don’t know why but I like java now. It is more verbose, means lesser chance of messing things up, no assumptions while writing. Thought that could mean there are more bugs since if we write more, it can mean more area for edit. But I like learning things, and Java is a bit challenging coming from Python.

  • The volatile keyword makes the variable read from the actual CPU instead of the cache. It brodcasts that the value has been changed and the thread has to invalidate the cache and later when it requires it, it fetches it fresh from the CPU.

Learnt

• Software can be made easily now, but that doesn’t mean people want to right?

  • Like the analogy from Simon Wilison on plumbing. You can learn how to plumb your house with youtube or learn to code by AI, write that software vibe coded. But if its really important to you, and don’t want any issues, you probably want to hire a plmber or a developer right?

  • Maybe the thing for software just breaks because no one cares what is underneath the software. But do we? If we use a pen to write things. We forget that there would be a person designing it, yes it was “made” in a factory, but its existence is due to a human’s thoughts and imagination, the care, the empathy for others.

Interesting Tidbits

• AI Map

Tech News

• Anthropic is no more Misanthropic, shakes hands with XAI/SpaceX, whatever

  • Yep, that explains this. Our enemies’s enemy is friend of mine. Anthropic and XAI have joined hands against OpenAI. This is AI Cold Wars.

• Cloudflare to lay off 20% of workforce, after Coinbase does 14%

  • Oh! This is another one. Not sure AI is the reason or is that a excuse for over hiring in the COVID era.

  • Whatever that might be, the term job security is a myth these days. There is no such thing.

• OpenAI releases 5.5 Instant

  • This looks like a flash model from OpenAI. If it was for free on the API, these might be my go to one. Haiku is expensive.

For more news, follow the Hackernewsletter (#793rd edition), and for software development/coding articles, join daily.dev.

That was it from the week 93. Time flies…

I don’t know what to do this weekend. Reading books, working with clankers, trad coding. Overwhelming is a thing in 2026 for developers. Whatever it is, you’d see something interesting in the week #94. The 2 year anniversary of this newsletter is 2 month away. Wow! 100 editions never knew I would complete them.

See ya next week, until then…

Happy Clanking ;)

Happy Coding :)

I enjoyed working this week, I am surprised I am saying that. The same thing a month back was dreading me, and now its making me wake up each morning. AI is kind of a progression and also a setback in some sense. I don’t know, but the usage will steer where the world moves with it. There will be a divide, of course, and if I am correct, it will create the biggest divide the world has ever seen.

But till that happens, just sip in the beauty of the nature, and read some handwritten articles, human voice, and some hot takes on the internet about tech.

Quote of the week

“To know that we know what we know, and that we do not know what we do not know, that is true knowledge.”

— Confucius

― Henry David Thoreau, Walden or, Life in the Woods

How true is that? If you know something, you are eager to express or impress, but when you don’t, you shy away or fear away. The later part is what makes us humans. The latter, the complete part, is what distinguishes us from the AI bots. They will hallucinate and pretend the opposite when they don’t know something, they are unpredictable even in the first known part.

Knowledge is not learning something, its about intuition and learning to adapt. Being honest is the best thing to do to yourself.

Read

1. To my students

   • Read this article, if you are a student or not. You will find answers and sentences that matter to your soul and not your ego.

   Above all, be motivated by love instead of fear.

   • Everyone is in fear, unsure of the future, don’t be. Recollect the feeling when you started to do the thing in the first place, I bet it was not money, if it was, and still is, then I don’t know, if it was love, keep on steering.

   Have the courage to go slowly, especially when everyone else is telling you that you need to go fast and cut corners

   • Vibe coding, running agents, and shipping without reviewing, these are hype and will fade with time. In professional development, at least, this won’t be the standard as far as I know.

2. Ghostty is leaving GitHub

   • This is bad. Oh my god, he has spent almost two decades on GitHub. I might not have lived consciously for that many years.

   • Man, this person has breathed and lived through GitHub. I can resonate with the line “When I went through tough breakups, I lost myself in open source... on GitHub”. I have done that to some extent, too. Not at his scale, but when I didn’t get any internships, got rejected in interviews, I found SQLC, Turbot, and a handful of organisations to fill my hunger for learning and programming. GitHub was the center.

   • I don’t rely on GitHub since I don’t maintain or do any hustles anymore, just because life is tough and so is the AI hype. I can relate to the outages happening these days. One of my projects flight-observatory, had a few failing actions notifications. I thought it might be 403 or some upstream external API issues, but GitHub was the reason behind it.

3. You can beat binary search: Quad Search

   • This is creative. Instead of just halving the list, we do a buckets of 16, quads, and compute which of the buckets the element can be. Since we have the max of each bucket. We can find the bucket and parallelly then compare each element in it.

   • Surprisingly, this is better for large arrays, and with multiple cores this will outperform binary search easily. Because we are not diving the array into halves anymore, we are picking the most possible region where the element could be and parallel searching it in those.

4. Full Text search with DuckDB

   • This is quite similar to what we do in SQLite. Pretty simple. I like how DuckDB just keeps on making transitioning to modern databases, but still keeping the backwards compatibility and keeping it simple and dead easy to use just like SQLite.

5. I got sick of remembering port numbers

   • He got sick of remembering port numbers, so he just reinvented ngrok for local?

   • Yes, it is different than assigning names to services, but I feel that it’s doing that similarly. It’s a clever trick.

   • This is the phase where people are building anything they want; the idea of personal software and developer tooling is just skyrocketing.

6. AI, Tokens and the gathering storm

   • This is quite evident from the recent events and rug pull of Anthropics subsidized tokens, and GitHub’s move from tier based to usage based billing. Its on the wall, its the end of the subsidized token era.

   • The next few months might actually tell us how expensive are these tokens.

   • Local LLMs be ready for shinny demo, the dark days might be about to end.

7. I asked Local LLM to add 23 numbers, I got 7 different wrong answers.

   • My last sentence was jinxed. LLMs can’t do math yet. Sigh!

   • I did explore this last year and they failed miserably, till date this hasn’t improved. I think we can create a skill for this shall we, I don’t want it to use a python interpretter for such a trivial thing to add. Can we make them add? It would consume a bit more tokens but let’s see. I have some ideas here.

   • But the point being, Local LLMs are not quite the hype that the propreitary models live up to.

8. Agentic Coding Fatigue

   • True. This is me reading to myself. Writing code gives us the clarity, becuase we were in the weeds, we know why each of the if else was written, we knew the edge cases, but now? We just run the slot machine to “fix it” prompt and cross our fingers so that it works.

   • The balance of understanding and the need to understanding, is wide and is growing faster than ever. I don’t know which one to lean onto. The former sounds like there won’t be any harm. But the software is such a term a thing, that nobody really cares how well you understand it. Only the output matters, not the usage of design patterns or Python or Golang.

9. Do I belong in Tech Anymore?

   • This one is really rough to read through. Not that it’s bad writing or thoughts, its is dead real and truthful. I myself am not able to express the true things that I have to go through, maybe I am not as privileged or have made my situation like this.

   • The principles that the author listed are gold, make sure we point it here too

     • Things that are worth doing are worth doing well.

     • Things that are done well require time and effort.

     • You make meaning through the doing.

     • Ideas are common; effort is not.

     • There are no shortcuts.

   • Frame it in gold, or memorise them, it is better to remember this than to be a slave to an agent or a corporation that is.

10. People who don’t use AI will get left behind

    • I will frame it as “People who use only AI will definitely get left behind”

    • People would stop learning and thinking if they just learnt to use AI and not anything else around it.

11. Have you seen the new Excel?

    • Learn Excel or get left behind, learn AI or get left behind. Everyone is on a roll with hypes. I do agree, excel has just revolutionized the way people think of software or even any form of work.

    • It has abstracted the code in such a way that people rely on the results and not the code. Superb product.

    • I worry AI is doing the same, but for text generation, for code generation. It might be the de facto thing to produce custom software on the fly.

    • Until that happens, keep learning Excel ;)

12. Where the goblins came from

    • This was an interesting nerdy read. I liked they are open about it. Though they don’t back away from shoe-horning codex glazing. I don’t doubt they use codex, but its quite a bit of token-hungry and slow model.

    • The problem of the goblin is really interesting. The goblin behavior wasn’t a bug, according to them, it came from reinforcement learning rewards that favored playful, creature-based metaphors in the “Nerdy” personality.

    • Those rewarded patterns spread and amplified across training loops, even outside the original context. It resulted in a small stylistic quirk becoming a generalized model habit, showing how local incentives can unintentionally shape global behavior.

Watched

• Making Sense of the AI hype - Wading through AI -Episode 2

  • The topic of engineer writing the blog and the manager or the sales person making the video is a reality and the only needle making the AI labs floating.

  • Its all gimmick play. Models aren’t quite capable of creating a novel thing like a compiler from scratch, that is, way far currently.

• AI isn’t taking jobs its worse

  • Anthropic is eating the money, by creating fear. Like organisations need things quick so they burn tokens, for that they need to cut down on cost, which surprise is got by firing employees. So, the money eventually flows to Anthropic by cutting the job, which is hype honestly speaking. Not sustainable, they themselves have proven it.

Learnt

• String Pool in Java

  • String Pool is a special memory area inside the heap where unique string literals are stored. The idea is instead of creating a new object every time, it reuses existing strings from this pool when possible.

  • A string in a string pool is a deduplicated storage of string literals inside the heap.

• Cursor agent > Claude code

  • I like cursor agent cli, it is intuitive and doesn’t block the user, Claude code is like I(the agent) am the owner. I don’t like that.

  • Cursor has the ability to view the output of the tool calls with Ctrl+o and let the screen continue, whereas the Claude just blocks the view.

  • I hate anthropic that is one reason. Maybe, but honestly speaking, claude code is blaoted.

Random Tidbits

• https://www.numberempire.com/

  • A cool webpage to see different patterns for a given number

Tech News

• Copilot is moving to usage-based billing: The end of a subsidized token era

  • This is a sign guys, if you still think, AI is not hype. The hype was at its peak, and now its wading. The subsidized token era is over. The real cost is on the labs and the providers that are now realising the importance of sustenance.

• Warp is now open source: Another attempt to win the race?

  • This looks like a desparate attempt to get into the lead of the race, I am not against it by any chance, its a solid move.

  • More organisations should do this now, espcially the ones that created this right? Yes I am talking about A…. Never mind.

• HERMES.md in git commit messages causes requests to route to extra usage billing instead of plan quota

  • Again. Never mind. Nobody would be using their coding agent.

  • They don’t leave a chance to not to hate them.

For more news, follow the Hackernewsletter (#792nd edition), and for software development/coding articles, join daily.dev.

That’s it from this week. It was a bit of a work-heavy routine. Finally feeling back, maybe not peacefully good but warmly good. Looking up to slowing down with manual coding and building some impactful things in the era of slop.

Till then,

Happy Coding :)

It was a comeback sort of week. I still kept learning Java and fundamentals. I drilled down on dependency injection, object internals, some react too in the mix. I found LLMs are good at summarizing things, some are good, some are hallucinating, some are just give up. Yes, and I have opinions in this edition about that.

The articles are a bit AI-pilled I must say, those are getting a little into my head, expect little or no article from AI next week. I had enough for the month. But yes this trend is not stopping. We’ll have to figure out to deal with AI now.

Quote of the week

“The difficulty lies not so much in developing new ideas as in escaping from old ones.”

― John Maynard Keynes

Its quite relatable here, Its not hard to develop new ideas now, we do it all the time, but now the problem is getting out of the mindset of living without AI or old mindset. It might not be completely AI but a mix of balance of conscious understanding of what to do and what to avoid.

Being a developer in this AI world might require us to shed certain ideas, the perfect code, the developer experience, the best design and all of that nitty gritty. It matters still but there seems to be no time for that, if LLMs are able to generate 10K lines in an hour, how can a human review that in a day? Its not easy, nor its impossible. But there would be a shift in this mindset of how we look at programming a few years back at least. And that change is very hard to make once it gets molded into the cortex.

Read

1. Lights-Out Codebases and the End of the IC role

   • Ouch! That one is brutal and eye-opening. Yes, looking at this, it just feels like software developer is no more a thing. But when I look at real world, nothing seems to be changed. Maybe its a bit slow, maybe people realize it slowly. But something is surely a divide.

   • Lights out codebases, this just makes total sense, I agree to that, because if LLMs just keep getting better, keep producing more code, more right code, human’s capacity to review that is out of hands. The chess engine analogy hits home. You don’t want to review the chess engine’s moves by a grandmaster right? Like what? See LLMs as the chess engine and the grand master as the developer. Yes, analysing a game between a human and a engine can be done with a grandmaster, but no one reviews code like a hobby. LLMs would react that point sooner or later where you don’t need to worry about the generated code, and at that point of time, why would one need a human developer to review the code.

   • Writing that passage got me some goosebumps, like its overthinking, but its grounded and factual. Not sure what lies for developers as a role ahead!

2. Software Engineering might not be a lifetime career

   • Maybe this is a bit negative, but factual, actually just tying up with the above post. It might not be slow, it might just wipe out the need for software workers in a single year who knows.

   • But world just changes drastically man, till 2025 people thought, software development is the hardest thing to replace with AI or robots, yet here we are the first ones to get down probably.

   • What would those people do? Farming, carpenter? use the software, yeah! we might become consumers rather than producers. No one knows.

3. AI Resistance is growing - read hackernews for better context hackernews post

   • The thread is civil but opinionated, with strong technical depth on poisoning feasibility mixed with philosophical debate.

   • Pro-AI commenters often acknowledge risks and call for better incentives, while AI-dommers focus on erosion of trust, job effects, and creative control, yeah! that is the typical conversations moving forward.

4. Developer Taste and AI Slop

   • This was an eye opening post. It is helping me understand, to differentiate generating vs developing. The code was and is never the bottleneck, what to build is. Even if that comes to you from other members upwards, a developer still has to think.

   • Anything can be made now, but if that is the real thing to make, worth it, or even needed is the question that would have an edge.

5. My dear friend you just built kubernetes

   • This one is quite funny. After resisting the complexity of kubernetes, people actually re-invent the wheel.

   • To use any tool or software, we must know exactly what we are trying to do and what that tool or software is actually doing. Not necessarily it would fit, but it would help you reach the ultimate goal a bit quicker.

6. Your company is a skill now

   • This s a bit scary. I talked about it when the open claw and moltbook thing was getting hyped. We were just a bunch of markdown files away from creating a dystopian world.

   • Markdown files are just instructions and context relevant for doing certain task and not doing the rest of the side quests. I know that sounds too simple, but that is what it is. But its not that simple. Putting ideas and limited context to words is not simple as you think it is.

   • Maybe creating or translating ideas to skills will actually be a skill in some months, who knows!

Watched

• Should you be a carpenter? Wading through AI - Part 1

  • Ok, the person who is into the weeds of AI models, himself is confused as to what to learn and what to choose as a career. This is a bit hard to swallow. Are we done then?

  • Nobody know the answer whether software is a safe career anymore. At this point, we need to define what a career actually is.

• I quit my GitHub Job because AI breaks software

  • Ok, that is positive but people can get jobs just like that? I think its quite rough out there.

• Microsoft accidentally told the truth about AI

  • Ops! Is it happening? AI Costs now trying to get better of the AI Labs. We saw with Claude subscriptions, now GitHub, cost cutting in place. First employees and now people will see the actual problem hidden underneath.

  • The tech debt after vibing everything. Maybe it won’t matter, because AI can clean it up, right?

  • But damm, the man just spoke a lot of things, they just gave me hope to be human. Just watch this, if you watch anything at all.

• Chai Aur Code - React Series

  • What a beautiful series. It feels nice to learn things. I never though React was that simple. It just injects html and js. Wow!

  • Props are just attributes that we pass around the html elements, this is just art of teaching at best.

• Did Anthropic just killed Figma?

  • Probably! This is a good product but not sure how sustainable it becomes. I mean I can use it if I have unlimited tokens and wallet to spend, otherwise its a big money burning machine.

  • And for that maybe figma is a good choice, old school. At this point, I am afraid what am I even going to see more from anthropic

• We need to talk about GStack

  • A bit of wasted hour I would feel. Never no major thing to notice, except just rambling. It was not worth the time.

  • Gstack is basically I think bash script to give skills to a agent. Nothing much, and maybe it helps you produce 40k lines of code a day. Who cares.

Learnt

• Dependency Injection with Spring Framework

  • This was a learning curve for understanding the core behind Spring Framework. Basically instead of changing values in object type, you just change it at the configuration level. So we don’t have to hardcode values. Its like the level of abstraction to define where to configure or inject the dependency. Hence the name.

  • Its quite clever to do this out of the box. In python I haven’t seen stuff like this. In django perhaps there is a way to just use settings.py and modify everything there without actually changing the model or view.

• LRU Cache

  • I tried to implement this in Java, and learnt to use Hashmap along with a linked list (doubly). Since we need to keep track of most and least recently used items, we need to be able to track it and more importantly for deleting and accessing in O(1) time, this structure just makes it quite great for this.

• Java equals and hashCode differences

  • The equals check if two objects are equal, their values and not reference value.

  • The hashcode method returns the hashcode of the object, its like the identity of the object.

  • The equals and hashcode needs to be consistent, so If you override the equals method, you need to make sure you keep the hashcode aligned with the equals too.

• React under the hood

  • Yes, I learnt a lot from the chair aur react playlist, and still learning.

  • React is a library, and Angular is a framework. React basically maintains its own DOM to effectively update the data.

• Table parsing for complex financial documents is still unsolved problem

  • Ok, I thought somewhere it might be solved but nope. Tables are a bit of a pain. Trust me when I say “pain”. A human himself cannot understand the layout, let alone LLMs. Poor LLMs. Even after giving them eyes (not realy) VLLMs are dumb too. I haven’t seen a great model like Gemini from other labs, if that breaks in, it should be a game over. But I don’t see those labs solving this problem.

• AI assisted/written articles needn’t be all slop

  • I just read this article sereis

  • https://nitinsingh717.substack.com/s/threads

  • Ok, please here me out. This is written by AI (but but). Its not slop. I honestly read it, all three posts. And it genuinely helped me understand concurrency better.

  • AI is not bad, but if you don’t have experience in the thing that you are generating, the taste (wow, everything is connecting now), then its slop, because slop is not AI-generated content, its AI-generated content that the author didn’t spent time thinking or reading about it.

  • And from the three posts I read till now, it genuinely felt that the model clearly understood the sentiment and the intuition of the author and it translated it perfectly.

  • I even discussed it with the author, and he admitted the same, humans are above LLMs for now.

Level Up Your Programming with Nitin

The Need: Why Sequential Thinking Breaks the Modern World

Series: Thinking in Threads | Blog: 1 | Read time: ~14 min…

Read more

2 months ago · 2 likes · 4 comments · Nitin Singh

• LLMs are a great tool for learning

  • Not all LLMs are equal, each has its own advantage.

  • I have found Grok to be best for search summarisation, Gemini for shallow explanations and quick searches, Claude for indepth reasoning and creative exploration. ChatGPT for general stuff.

  • Claude models are just great for agentic work, but they don’t quite search well. Grok is reallly great at searching and getting factual data in the context. These are opinions and not facts. But I have used all four enough and continue to do so, knowing what each of one can do and where I should throw which task.

Tech News

• OpenAI release GPT 5.5 - doubled the price of 5.4

  • Another release. So, its plateaued now. Models are stagnant with the knowledge, and only breakthrough would be to make it sustainable for inference, rest is looking RL and post optimizations.

• OpenAI releases ChatGPT Images 2.0

  • Oh! This is good. Like the level of detail on the image I generated for Java OOP was just top notch. The text is just flawless.

  • Nano Banana was not that good at that precision. The crown has been stolen from Google to OpenAI again.

• Canonical releases Ubunutu 26.04 LTS - Resolute Raccoon

  • I just installed it on my work laptop and it was flawless. It has no major changes, but just software upgrade. However more rust is starting to get into the linux core.

• Kimi Releases K2.6 - Open Source model are still keeping up with proprietary models, a good sign

  • Wow. Open source models are not dropping the knees. The fight is on, the open source models are not as accurate or reliable as closed source but they are getting closer. The weight is a big factor for their adoption, if they can squeeze a great model in a relative smaller size, it can be a win for those labs.

• DeepSeek releases v4 preview with pro and flash

  • The fight is on. It keeps on pressing the closed source models to get out better models and we are stuck in a infinite loop.

• Anthropic still shooting self on the foot - Mythos is over-hyped

  • Yeah! Anthropic is not leaving their steak of shooting themselves on the foot. It has been over a month and the saga continues.

For more news, follow the Hackernewsletter (#791st edition), and for software development/coding articles, join daily.dev.

That’s it from this week. It was a back in track, old vibes week. The next week is a crucial one, a make or break situation for me. If I am making a change, this is the moment. Its going to be hard, but I think I can give my best. The worst has already happened, there is nothing that can go worse than this situation for me. Anyways, looking forward to a positive week ahead.

Until then,

Happy Coding :)

Week #90

Another rough slog of a week. But fighting nonetheless. Hopes are the glasses we have strapped in, darkness cannot blind us.

I have been upskilling myself with a few things, locked in with handcrafted coding. I am spending at least 2 hours in the morning to solve problems without chatgippty in my way. I don’t use agents and no AI. I think there needs to be a restraint in building and improving code skills. So, planning to do weekly livestreams for that public building and tracking my progress, also it helps me be accountable of what I say and commit to.

This week I am learning Spring Boot and Java skills, I don’t know why, don’t ask me yet. But I don’t learn anything without reasons, it might be curiosity, it might be a requirement. Learning is always a skill of a developer, and each year I want to learn something different. Maybe 2026 would be a year I would learn Java for enterprise-grade? Wow, that will be a great flex to make while people are prumpting with their AI-buddies.

Quote of the week

“What I cannot create, I do not understand.”

— Richard Feynman

This is my mind right now. If I cannot hand code it, I don’t understand it. Your counter point might be, no one cares if you are anyone understands it or not, its code, not a form of art that is valued. Yes, that is true, users don’t care a damm if I hand coded it or the clankers generated it, but when it breaks, who is responsible. Yes, say it. It is a developer. No one can blame the clanker, people exist to blame people. I am not saying to blame it or not, but to take action and a decision from there on to resolve the issue, is only possible if I know the ins-and-out of the program, that only happens if I am in the weeds of it. Reiewing code can do it, you think, sure. Do it, and don’t complain after 3 hours of glazzing on the clankers’ 3k line of code slop.

I need to create in order to understand, I need to write it to know what it feels like to break and make. Before reviewing the code, I need to know how it works and what can break.

Read

1. Frustration Driven Development

   • Wow! No engineering is good engineering. Remove the problem from the root, well said.

     Your job is not doing the work, your job is removing work.

   • How hard does that hit? In the AI world? Automating our own job, yeah, that’s fun, right? Maybe!

   • But this is a great point, great software engineers use swear words and get the problem out of the way, and do not work around the problem. (as codex does, and Claude code just wraps around the problem when it sees it)

2. Who will be the senior engineer in 2035?

   • Totally valid point. If AI is becoming the cheap junior dev, who are we upskilling for the future?

   • The gap is already wide. In a couple of years, it might grow like crazy, and the divide of expectation and reality might hit people and developers alike.

   • There is and will be an expectation of in-depth understanding of code, but by using the AI-Agents, the quality will deteriorate drastically; developers won’t read and write code, and the instincts and the muscle memory to write code without assistance will be gone.

   • I am thinking of doing weekly streams now, to hand-code certain things. Maybe that will keep me up my ante of vim flexing and away from vibe coding.

3. Lorin Hochstein’s thoughts on BlueSky public Incident writeup

   • That is a brutal one. This is like concurrency, you thought you had one problem, but after adding concurrency, you now have two. Setting a limit on how many go-routines can be spawned in a group is critical here.

   • TIME_WAIT, that is really a neat thing to learn. The TCP connection waits for that duration before sending another FIN so that the client can be sure of the delayed packet delivery, if any. And that TIME_WAIT actually caused them to fill up all the ephemeral ports. That is 28k ports, which sounds a lot, but after reading it through, that is surprisingly a low number. If you are Bluesky scale, you might need millions of ports.

   • Diagnosis skill is something that is going to be super valuable going forward in the AI era. AI can help, but it would be too slow and can never reach the instinct-based debugging of humans (as of now, at least)

4. Six Characters: Decoding PNR Number and e-ticket system (Ajitem)

   • This is a treat. People should write these kinds of blogs. This is curiosity at its best.

   • PNR is not a unique global identifier, it’s specific to the airline or the entity handling it, it’s for passenger name record.

   • It’s quite ingenious how the currency conversion works, without breaking and keeping simple and straightforward.

5. AI will be met with violence, and nothing good will come out of it

   • Spicy take. True in most senses.

   • The CEO things are pretty messed up, people might fight and get violent, which is inevitable in any direction we go.

   • The start of the post was quite well written, if that holds true, we can actually see where this will go.

6. I’ve been blogging wrong

   • People are waking up and finding the strength to write authentic content. Sharing human experience, which was the sole purpose of blogs. But it has taken the shape of vanity metrics and technical jargon. We are so back!

7. Llama Parsebench: First Document parsing benchmark for AI Agents

   • I am a bit surprised that there are no document parsing benchmarks yet? What? Where are the Chinese labs and Msitral and all the OCR benchmarks then. Oh, they might be just OCR is it? Well then that makes sense. For document-specific parsing.

   • Another industry or field demolishing with the report card calculator ready.

8. Now is the best time to write code by hand

   • Yeah! People are firing now. This is happening. Writing code by hand can be a hobby but it will be a skill that pays like COBOL or PASCAL developers are paid today. Trust in code will be more from human than in agents, that is the bet we are making if that holds, software developers are going to be PHP developers with Lambos in their garage.

Watched

• Data Structures Explained by Nic Barker: HashMap

  • This was a solid explantion of a hash map. I actually thought it was magic, but I knew it was some hash function or hash code, but this actually clears a lot of those magical things.

  • I learnt that we can have linked list sort of a structure to tackle collisions on hashes. That is a really clever way of solving a problem.

• You are still better at editing documents than AI

  • This is a great explanation of why the software interface is skewed for agents. It was never meant for them.

  • The right set of tools and context is really important. It might also depend on the model, since its training data, if it doesn’t have the right examples on how to manipulate docx or xml files, it might skrew up big time. But I have not seen those kinds of issues from proprietary models yet.

• Vim has a 0 day?

  • This is evident, and quite not surprising to me. AI agents and slopy review system will make this worse.

  • Vim codebase was hand-chiselled for more than 30 years, it still has these vulnerabilities, now take AI into the game, and it could easily make it worse in a matter of months if not years.

  • Reviewing code and testing it will be quite a skill to have. Security essentials will be key in moving out of this kind of mess. I also read about the diagnosis skills in the read section, right from the Bluesky incident report, so that will be another skill to hone.

• Anthropic Dunking and PI pilling

  • Oh! That was a pure entertainment. Grilling Anthropic is my new pastime hobby, or I should say its Theo’s duty.

  • Anthropic is reduced to Syltherin and worse, for some time last year I was considering them Gryfindor, oh my god, how idiotic I was. I thought they shared the safety scores, and made it specific to developers so it might be nice.

  • But the CEO and the company are some evil propaganda to get funding. Good luck getting rejections.

Learnt

• Difference in this and super in Java OOP

  • I understood that this refers to the current object instance and is primarily used to access instance variables, methods, or constructors within the same class. super refers to the immediate parent class and is used to access overridden methods or parent constructors. The important learning was not just usage, but when ambiguity arises (e.g., variable shadowing or method overriding) and how these keywords resolve it.

• Internal working of HashSet and its relation to HashMap

  • I realised that HashSet is backed by a HashMap, where elements are stored as keys and the hashvalue as its value. This means all properties of hashing, collision handling, and performance characteristics come from HashMap.

• Design Patterns from Spring Boot

  • While studying Spring Boot, I learnt that Dependency Injection (DI) is a way to provide an object with the dependencies it needs from the outside, instead of the object creating them itself. This separates what a class does from how its dependencies are created. Without DI, a class directly creates its dependencies using new, which tightly couples it to specific implementations. This makes code harder to test, extend, or replace. DI removes this by depending on abstractions instead of concrete implementations. Its pretty good pattern to be honest.

• While brushing up on concepts on OOP, I also learnt about interfaces and abstract classes. An interface defines what a class must do, and an abstract class defines what a class is, along with some shared behavior.

• The software industry is in extreme ends

  • On one hand, we have AI-pilled managers and bosses. On the other hand, we have Java coorporate coder (hand-crafted coders).

  • Like if someone looks at the job market, it’s a chaos of expectations and reality. How can a field have 3+ years of GenAI experience when the field itself is publicly known to be less than 3 years old?

  • I must say, not everything is doom and gloom, there are still companies expecting people to hand-chisel code, and that is where Java enterprise comes in. Maybe that can change my trajectory, the next two weeks are the decisive ones.

Tech News

• OpenSource is in danger

  • This is bad. Yes, we had around 5 supply chain attacks in a period of 2 weeks. That was brutal, but one cannot blame it to open source, and it cannot end.

  • It is the pillar on which the tech giants and the smaller firms are standing on, it cannot be something that you go private for. Its shows the lack of trust on the 80% of the people doing good things, but those 20% of the people yes that is brutal.

• Google releases Agent CLI for building Android Apps

  • This is actually something useful. Opening Android Studio is like baking a grilled sandwich. I have to open and close my laptop 20 times and cannot build the app.

  • With the agent and the tool chaining, this could help me produce a slop of apps.

• Anthropic drops Claude Opus 4.7

  • Oh another 0.1 bump. Nothing much. This is fine, we get impressed, we suspect, and we reject it after 20 days.

For more news, follow the Hackernewsletter (#790th edition), and for software development/coding articles, join daily.dev.

That’s it from the heavy Java-pilled edition. Never thought I would be learning java, but here I am. never thought code can be generated that correctly and that fast, yet here we are. Life has a lot of surprises for us, so don’t get discouraged, you might not know what happens next.

Until that happens for good, keep your hand on the keyboard and brace yourself with the nerves of reviewing code.

Happy Coding :)

Leave a comment

Share Techstructive Weekly

I did a few things this week. I made a little app, wrote a half-baked draft for part 2 of the flight observatory, and learned design principles.

But something feels off.

I am feeling a bit rusty with programming lately, not sure where the joy in coding is lost. I hate managing agents, I hate reviewing 2k lines of code in 2 hours. It’s just not what I had signed up for. I am complaining, yes, but it’s just a weird thing.

Not feeling great to be honest, have ideas, but agentic-engineering seems to have completely overtaken. I am saying that I miss writing to code by hand, or I want to, but it feels like the layer of agents has replaced that muscle memory of thinking while coding.

Anyways, I still find the joy of building, not sure how long it will last. At least can be hopeful about it.

Quote of the week

“We are often most in the dark when we are the most certain, and the most enlightened when we are the most confused.”

— M. Scott Peck

Maybe, not sure. I feel I am up to something. Maybe the path is clearing up; the night is darkest just before the dawn. Keeping a positive attitude is the way forward. I think I have a lot of things to juggle and grasp, which is why I feel lost. Once I can manage the right things in the right direction at the right time, I can feel a bit confident. Cannot see the other way round. Escape is not an option, and I don’t like it either. There is no fun and learning in choosing the easier route. Ending with another quote, easy routes, hard life, hard routes, easy life.

Created

• Chessight → https://chessight.meetgor.com/

  • I built it as my experiment to learn agentic-driven development, of course, why not! I used Codex, and it just one-shotted the entire UI, I deliberately asked it to use Vue, and it did it phenomenally well.

  • Ok, what is the project? It's just a visual training application that can help you understand the chessboard and its algebraic notation.

  • The idea is simple, not to teach chess, but to train how you see the board, like g5 shouldn’t be something you calculate, it should just click when looking at the board.

  • I was watching FIDE Candidates 2026 (Prag vs Sindarov), and I was intrigued by Sagar Shah and all the other chess players and commentators who can just name the moves. I was still figuring out the move, and they named the exact step. I wanted to build that muscle memory. I looked for chess.com and Duolingo, but neither of them provides this. They don’t start from the fundamentals.

  • After playing it for a while, it feels like one of those small skills that unlock a completely different level of thinking in the game. I am not a chess expert, I don’t play chess regularly, but I like it, just for fun and flexing the -100 IQ brain of mine.

• Draft for Flight Observatory technical report and devlog.

Read

1. AI and Remote work is a disaster for software engineers

   • Spicy take. I’ve been working remotely for ~2 years as a junior-ish backend engineer, and I use AI daily. If AI + remote sabotages juniors were broadly true, I should be a pretty weak engineer by now. That hasn’t been my experience at all.

     remote work sabotages careers of youngsters

   • I think remote doesn’t sabotage. It removes forced structure. If you don’t build your own, you stagnate and just feel lost. There’s no overhearing seniors, no accidental learning, no pressure to “look busy”. If you just do assigned tickets and log off, yeah, sure, you’ll stagnate. But that’s not a remote problem; that’s an ownership problem. You can sit in an office and do the same thing.

     AI — too little brain stimulation

   • Same with AI. If you use it to skip thinking, you’ll get faster at producing things you don’t understand. But if you use it to explore, ask “why does this break?”, generate edge cases, compare approaches, it’s like having a patient senior who will walk through things with you endlessly. The difference is in how you engage with it, the intention rather than the environment.

   • I do agree that juniors need tight feedback loops and exposure to better engineers. That’s harder to get remotely, and most companies don’t compensate for it well. But the answer isn’t “go back to office”, it’s “design better learning environments”, more deliberate mentorship, better code reviews, more context sharing.

   • Also, the market point is real: AI is eating the bottom layer of trivial work. But that just raises the bar; it doesn’t remove the path. Juniors now need to show they can reason about systems, not just implement tickets.

   • The point is, if an individual is naturally curious, remote or on-site doesn’t matter, he’ll succeed in whichever environment.

2. Someone at BrowserStack is leaking mail addresses of customers

   • This is all happening. The security of software is crazy right now. We just had a week of supply chain attacks on Python and JavaScript ecosystems. And now its getting into the weeds of the application. Not too far from AGI, right?

   • I don’t reckon that the company might be leaking it intentionally, it might be some vendor they aren’t fully aware of, or have really vibe-shipped something. Not sure.

   • Excited for the next post.

3. Does coding with LLMs mean more microservices

   • Yep, this is a valid and good observation. I have mostly written my side projects in cloud functions (Netlify, Vercel, or Cloudflare). I don’t mind the complexity of managing a bunch of environment variables and setup. It’s a one-time thing, and I know looking for them is a bit of chaos, but I have AI-agents to clean up as and when necessary.

   • Maintaining a monolith is no joke with LLMs, it used to change hundreds of out-of-the-blue things, nonsensically, but it has reduced almost to 0. But the cost of changing one thing is still high in those environments, since a change without review can cause a catastrophic production failure. (Just like GitHub is having right now)

4. Software never had a soul

   • Beautifully put.

     “You do not even need better or faster tools. You just need to really mean it.”

   • You need to mean it, you need to care it, this is the crux of being a developer. We should not be caring about the code but caring for the user’s problem, the product.

5. People love to work hard

   • Great point. This is about not what you do, but how people empathize with your work. If they don’t they say “you don’t want to work”, if they do you can see the title of this post and read it.

   • A very human post from this author. Kind of heart-touching. There is nothing more to say, like it’s something you feel and can’t describe in words.

6. Lessons from using SQLite in production

   • Useful bit here. The WAL mode is absolutely a clutch, most people don’t know of, and just lament about it being a single-writer constrained DB. It’s a super-powerful and very versatile lightweight database.

   • Folks at Turso are making it even better with read-write replicas and syncing, having a daemon for SQLite. The future might be apps full of SQLite.

Watched

• Simon Wilison on Lenny’s Podcast

  • This was a relieving video. I thought I was the only a minority amongst the developers who got exhausted after a few hours of AI prompting. It is a natural thing. We are doing the creme layer, the ephiony of the task, this won’t be sustainable.

  • The other thing is that he had more than 2 decades of experience.I am not giving excuses, but still, that is a lot more than just a 2-3 year me trying to wrestle the concept of agentic engineering.

  • It was refreshing to see this, his enthusiasm is truly contagious. I want to build more now. But mindfully. His tools section is really wild.

  • I fit in the middle, I not a junior or a fresher anymore, I am not more than 2 years into the industry. Alas! I am in the middle. But still I think, its not doom and gloom, I am not going to be passive and let the AIs take over my brain.

• WhatsApp System Design By GKCS

  • Well explained. I thought it might be more complex or elegant. But it was neither. Its a chat app nonetheless that scales, that’s it.

• Building Zepto LLD, System Design

  • This was a good design. The system feels a bit chaotic, but was understandable.

  • The inventory, delivery, product, and everything could be a system on its own, but in the real world, systems are created by combining multiple systems.

  • Basically, we have products, orders, and order items. The inventory manager finds the product in a dark store(large storage or inventory) and then assigns the nearest dark store with those products or multiple stores, and maps the available delivery agents to those shipping those products as orders.

• How UPI Payments work

  • Oh, so every bank, small or national or not, has to go through a nationalised bank(or one of them), damm!

  • I wonder how and why that is so fast? Is it because the banks already have mapped out the route to each bank amongst them, yes, that’s probably the reason for the unified name.

• Strategy and Observer Design Pattern

  • Banger of a series. It was a great explanation, and it felt really enthusiastic.

  • Strategy patterns are like polymorphism, but for a specific component or functionality. Like payment can be a strategy pattern, where we can implement UPI, card, net banking, etc.

  • The strategy pattern is like implementing a family of algorithms and encapsulating the algorithms in their own, and we can make them interchangeable.

  • Observer pattern is like a push strategy rather than the client polling it. It makes it efficient and clean. When the observable changes, it notifies (pushes) the changes to the observers, and then it can fetch the changes since it can decide what it wants.

Learnt

• System design is a really broad field

  • I thought it was a bunch of arrows and squares. But its a little more subtle than that. It’s about abstract design, and schema too, without going into the code details. That’s a bit tricky part, because creating an abstraction not low-level as code and as high as a class or an entity is a bit tricky balance to maintain. You’ll either go all in or barely touch the core problem.

  • There are design patterns and principles to use, diagrams for them, UML, ER, HLD, LLD, and whatnot.

• Design Patterns are wired, but important

  • This came as a bummer in an interview. I knew patterns, but couldn’t recollect their names, and was dead on the lines. I wonder if people actually care about theory in the age of LLMs. It has made learning on the fly so easy that I am starting to not learn anything deeply. That’s a hard realisation to come out of.

  • This came to me, and I immediately started learning and reading about, I crunched almost halfway through “The Object Oriented Way” by Christopher Okhravi, and referencing the “Head First Design Patterns”, the mammoth of a book by O’Reilly.

  • I am finding it rewarding now to go through them. Really, LLMs might make us the dumbest creature on earth one day. The greatest devolution in the history of Earth.

• We are in a software boom, and not doom

  • I am realizing that there are more software engineer roles open, more need for software, but not quite one-to-one. It’s like people can make software that they once couldn’t, but now can do a bit easily. But the only thing that keeps them holding back is that they or anyone doesn’t know what the code actually does. And I think as software developers, this edge will be vital more than ever.

  • Till people cope with it, it’s time to go deep and explore the depths of software.

• You can make your own small LLM

  • This is very cool. I like it and found it almost at the right time. I wanted to build an LLM with some specific data, but I was not sure what that could be. This guy just created a bunch of fish and food sentences and called it a day. Legendary stuff.

Tech News

• Anthropic with Claude Mythos, Project Glasswing, eh?

  • Maybe it was a great mode, maybe it was a meh! model. I don’t know since I don’t have access. Period.

For more news, follow the Hackernewsletter (#789th edition), and for software development/coding articles, join daily.dev.

It was a slow week. To be honest, disappointing in some sense. Not much movement or progress. But the next week might just be good or great. That’s the hope as usual.

That’s it for this week.

It was a high-action-packed week. I wrote a case study, polished a web app for that, and connected with people for regular catch up. Since job security is not a thing, as we can see with Oracle. Negativity aside.

I was really excited after the weekend, since I had all the queries and data for getting my intuitions and memories right on the Mumbai airspace case study. I wrote the blog post, which took some time; to be honest, it was hand-chiselled. But the feeling of completing it was something that nothing can come close to. It was a small thing, but I might be exaggerating it since it was a lot of work. I took 2 weeks to build it. Research on the data sources, download, and fail to load the data (it was a mammoth dataset, even sampled), tweaked and changed the architecture again and again after surprises, and finally made sense of it. There is nothing more rewarding than that.

Apart from that, I am also planning to do some live streaming, maybe for upskilling my coding practises. I am with Primeagen here, he was right that we need to have a moment with ourselves, to resist that urge to let AI handle this tiny thing, and it is infatuated with AI doing everything. I want to have my mind disciplined in writing code by hand. Maybe the future is AI writing code, but I cannot understand what it has written until I feel and write some of it myself. I might be stubborn, but that is what it is.

I read a lot of good bits and pieces this week, also I don’t watch youtube as I said for the past couple of months, it's just AI-like slop and reaction videos all over. But some of them are gems too, I admit.

Quote of the week

“When everything starts going against you, remember that Airplane takes off against the wind and not with the wind”

— Henry Ford

This is used intentionally since I made a flight project this past week or so. I wanted to relate to it, and it hit me at the right time. I didn’t breeze through the project; I faced issues that, as a developer, we have to. I learnt a bunch of things and made it through. There will be resistance in moving ahead, but that doesn’t make you stop; it makes you value that even more, it fills you with empathy and experience.

Nothing in life will be easily attained; it might be, but with subtle or bigger hurdles. You have to learn to fly through it.

Created

• Flight Observatory: Finally launched. After 2 weeks of fighting with data, it is here. I also re-designed the home page and the archive page. The webapp now displays the last fetched status for ads-b live data and archives it per hour (if the github cron job action runs on time, it writed that it doesn’t).

• Mumbai Airspace Case Study: The blog post and the case study webpage are live. It was a great one, I think the post is my all time high in terms of words (8k). I found it really enjoyable to write it, to finally validate my memories with actual data.

Read

1. I used AI, It worked, but I hated it by Taggart

   • A banger and the best blog I have read in 2026. Period. Just read it. Please give that author a hit on his site for his work.

   • But I will write about it anyways. It was a honest and blunt post, but written with both sides in mind, very balanced but favouring the other without defaming other side. Just take my money for writing this post.

   • I have been screaming this, but nobody realises it

     QUOTE ”the tool requires expertise to validate, but its use diminishes expertise and stunts its growth. How does one become an expert? “

   • banger after banger in the post, still saying you read it, please!

     QUOTE “I turned to generative models not only as an experiment, but out of desperation. I had a need for code that did not exist. Nobody was going to help me build it, nor should I expect help for a project.”

   • How many of you fellow developers are feeling it? True and very resonating

     QUOTE ”For any new potential project, there is a voice in my head telling me how much easier it would be to let the model do it”

   • Sigh! this is sad part, we can’t box it again

     QUOTE ”If I could disinvent this technology, I would. My experiences, while enlightening as to models’ capabilities, have not altered my belief that they cause more harm than good. And yet, I have no plan on how to destroy generative AI. I don’t think this is a technology we can put back in the box. It may not take the same form a year from now; it may not be as ubiquitous or as celebrated, but it will remain.”

   • Great post. I am just processing it all. The words just hit hard and then resonate perfectly with my experiences.

2. Resilience in the age of AI

   • This is a spicy prediction. I think this is a far dystopian prediction, but you never know. I had thought software creation was a far far dystopian fantasy, but here I am experiencing it. In late 2025, I thought it was a decade away, only to realise it was a few years away and then to wake up and realise it was already happening. Damm! This AI Era….

   • The thought about privancy is really going to happen or is already happening with the dangerously skip permission or yolo modes in the code assistant terminals. This will spread in “whatever you want, take it”. This thing will test the security and design of the AI systems that we will be producing.

     QUOTE “While today there are concerns about personal privacy and security, in the future we will be much more willing to share information about ourselves to avoid ambiguity in our requests”

   • This looks feasible and could happen. They will resist it, everyone will as usual. But the thing that scares me here is there would be no linear trajectory of careers. Like if you work more harder you won’t be rewarded as much. Since AI will do the most work, you’ll just manage them, I don’t know what I am even talking about. But the trust and credibility will blur with these systems for sure, or atleast will be hard to earn as it is today.

     QUTOE “We can break down future employment categories into three major branches: those who care, those who service and those who experience”

   • Well concluded. If we need any of these, AI has failed us, humanity has not failed.

     QUOTE “If there is an ongoing need for leaders, educators, financial workers or professionals, this will be a sign that the AI revolution has ultimately failed and will signal a long-term limitation in the aspirations of humanity as a species”

   • I wonder what resliance meant in the title, I thought I was going in for a full resistance to AI, but found out the oppsite. So it would mean resistance or resliance to your older believes no longer will hold in the AI era I suppose? But older experiences might.

3. Don’t let AI write for you

   • 100% Believer in this. If you can’t write the words, you are letting the world take over you. I just wrote a 8k words, though it was not 100% by my mind, I used some inspiration to talk and extract ideas out of my results and intuitions. But I wrote the full post myself, word by word. It gives a different level of satisfaction and authority that no LLM can.

   • It also is important aspect to build trust and connection, even resonance. LLMs are blunt and boring to talk to. When I write something, I have memories and thoughts that LLM can never have. That level of detail, it can have, but the emotion and the right set of words, LLM would never take off from humans.

4. I am definitely missing the pre-AI writing era

   • Its actually a great post. I cannot relate on the writing side, I can always rant on for 2k words non stop just like this newsletter. But I can relate this to AI-assisted Coding. I think I might be a bit rusty if I had to hand write code now. That is something I wish not to do, but the industry is forcing the other way.

   • I never use AI to touch my words, I just use Grammarly to refine the word mistakes, that’s it full stop. Nothing ever touches the world. I throw it to ChatGPT, Claude, and Grok turn by turn to rate my writing, and I take some critique from them. I use the extra “be blunt and brutal but honest to rate this” to add a negative direction, forcing it to find mistakes. That is a good use of AI to improve your writing, but not accepting it blindly and pasting what it throws at you.

5. AI overly affirms users asking for personal advice

   • This is partially true, as I said in the above thought to add a “blunt and brutal but honest and grounded” advice, it steers the other way. Its not a sycophancy but more about instruction-following.

   • ChatGPT models are increasingly becoming instruction-following, but are overly sweet sometimes, I agree. Grok is the other way, as we know from the snitch bench.

6. Software is a feeling

   • I can see the frustration there. Maintaining a blog is wired. You want to have one thing, then another, and another and it just breaks the full design and flow.

   • Great blog design btw, it has inspired me to have my blog in a VS Code-like interface, just wondering and being a little more ambitious than I am, because I have LLMs to design it ;)

7. LLM Knowledge Bases

   • This is neat stuff. I don’t have any wikis like that. I haven’t explored something that deep that requires more than 10k lines of content. I think I need to read one deep posts or a book. I do read some short-form content like other people’s thoughts and TILs, which is not bad but not really putting my brain to think.

   • I have seen Grok do really good extraction of information (not presenting the right way, though its not a great model for conversation, I suppose). LLMs are actually getting good at summarisation and linking different ideas together. Some times they are a bit cringe and try to shoehorn some weird analogy, which I have noticed.

8. What is this job anymore

   • I think I am a bit confident to say, we are a few months away from non-technical people saying, “oh crap, this is seriously bad!” (probably some security loopholes and ridiculous architecture) and saying we need some person that can fix it, the guys who are paid to edit text you know those, nerdy people out there. I thought our job was gone, we will be so back.

   • Back to not so sarcastic talk, I agree to the post. I am constantly asking what the hell it means to be a developer now? Sure manage agents and their output, but we weren’t build for that were we?

9. Relaunching the instapaper API

   • Saw this coming. Was doing development on this in September-November 2025. But alas! Some procrastination habbits never change. I wanted a reader that can just help read without any distractions. Just text. And this instapaper is doing just that, wondered if I can make something like it but more robust.

Watched

• Claude Code sourcemaps leaked

  • This is dramatic and really wasn’t needed, if they had opensourced the Claude Code in the first place. Like its now open atleast once, people have reproduced it in Rust, what is the point of hiding it?

  • This is kind of idiotic from Anthropic. I don’t know what the “secret sauce” really is.

• Sapling

  • This is a cool abstraction for git CLI users. I like the prev and next for branch switching and forget for rm —cached moving the file from the staging back to unstage, really neat and intuitive to use.

  • The web is also nice, the split is really handy in the times of AI if anyone is reviewing the code and you want to have mercy for them ;)

  • But the undo feature is so great. Like that is some super power of git hidden behind some awkward commands but that interface just made it a piece of cake. Amazing stuff.

• Analysing YC batch with AI and reinventing how to plot charts

  • This is typical twitter and linkedin these days. Slopify. I hardly read posts from twitter or any social media directly these days, only if I am following him or her then only I know its an authentic post. Otherwise its a chaos to find authentic posts from the slop.

  • The video is an excellent example of that. AI is everywhere that is the conclusion.

Learnt

• JSON is worse than CSV, CSV is worse than Parquet. Parquet is the best

  • I also learned pretty quickly that the file format matters a lot. JSON is flexible, but for analysis, it is awkward because it is nested, noisy, and expensive to keep parsing over and over.

  • CSV is easier to work with because it is flat and tabular, but it is still just text, so it does not preserve types very well and gets bulky fast.

  • Parquet is the one that actually feels built for this kind of job: it is columnar, compressed, type-aware, and much faster to scan when you only care about a few fields or a few columns. For DuckDB specifically, Parquet was the best fit by far.

• Actual use case based learning for Windows Functions

  • While writing SQL queries on duckdb for analytics on the Mumbai airspace case study, I wrote some terse SQL. Like some of them very 50-100 lines long (not full lines, formatted i mean) but they were long like a train.

  • I actually understood the need and the intuition to use LAG and LEAD pretty well. Partitioning looks easy now, though I need to be very careful about what and how.

  • I used LAG for detecting breaks in a flight run. If altitude, speed, callsign, or timestamp jumped, the current row was probably the start of a new movement.

  • I used LEAD for finding the next landing, takeoff, or phase change. That is how we can measure gaps like same-aircraft turnaround time and spacing between consecutive events.

• Technical things on Mumbai Airspace case study

  • What really clicked for me was that Mumbai’s airspace is not “busy”, but the reason behind it gave a more satisifed answer. Once I took the raw ADS-B rows and turned them into actual runs, the shape became obvious: the same corridors kept showing up, the arrival side stayed in the airspace longer than the departure side, and the whole thing looked tightly constrained rather than random. I had to use sequencing logic, especially LAG() and LEAD(), to see the transitions properly. That’s what let me detect where a run was changing, where a landing really began, and how quickly the same aircraft came back out again. So the interesting part wasn’t just the totals, it was the structure behind the totals.

  • The other thing I noticed is that the mix is very concentrated. It’s not a broad spread of aircraft and airlines doing evenly distributed things. A few narrowbody types and a few major carriers do a lot of the visible work, and that makes the airport feel like a high-rotation system inside a very tight urban envelope. So the data story for me was basically: once I cleaned up the noise of telemetry, Mumbai airspace looks like a constrained operating pattern, with strong corridor reuse, quick aircraft turnover, and very clear operational pressure.

Tech News

• It was a week full of supply chain attacks, litellm then axios

  • It is somehow becoming a trend in 2026 isn’t it? Supply chain attacks on code packages?

  • Axios was the latest victim of it. We are just past the first quarter of 2026, any more to come?

    And then it was

• Anthropic DMCA run, they were dmca-ing repos they themselves have leaked. Really they have made a mess of themselves.

• Google releases Gamma 4

  • This actually looks kind of neat. I tried the smaller base model on Collab and it was somehow a good starting point. Maybe the instruct model would be better, but need to verify with the capabilities and the knowledge it has.

• Cursor launches v3

  • It still feels like an IDE though. I am not complaining, I am happy, so that means they still believe that AI can’t work without developers, people still need to look at code? Maybe, just saying and thinking out here. Because cursor feels and is an IDE which means developer needs to be hands on with the code still.

• Oracle Layoffs 30,000 employees (I didn’t find the official source

  • This is brutal. Despite being in profits, why would one firm fire off 30k employees? That is a trash decision or dishonesty.

  • I have read posts from people laid off at Oracle in this wave which were serving for more than 10 years, 16, 33 years. Like goodness! What on earth have gotten into the companies?

  • I can just wish, these people find a stable and peaceful life soon.

For more news, follow the Hackernewsletter (#788th edition), and for software development/coding articles, join daily.dev.

Phew! Another week down. This one was a great one. A lot of things are finally moving. The tech situation is kind of messed up right now, but it seems there are more problems to solve than ever. Let’s not use that word for the ending now. We can still build projects and write about them as we did before in the past before this era. We can just stay curious and still breathe, I think that what matters as humans, I guess.

That’s it from this one, see you next week!

Happy Coding :)

Week #87

It was an active week. Did quite a lot of stuff. Most of which might be wasted, but lessons learnt. Pulling datasets in GBs from the internet is not a easy task, especially if its in chunks of JSON.

I completed CS50 SQL course and created the final project linked in the created section. It was fun, learnt a lot. I had watched the full video series a few months back. But never got the full time to complete the problem sets and the project. Locked in the past weekend, and completed the full 7 problem sets.

Flight Observatory Project

I also worked on creating and making the flight observatory. Its a bit of a challenge, which I want to complete by hook or by crook. It is a dataset issue, the data is there in samples but that too is quite too much to pull off. I cannot find a easy way to grab all the data for Mumbai only region. Loading full is not feasible and cannot manage it locally. So, loading each file in memory and filtering based on positions and saving the relevant bits into compact CSV is the best I have gotten, but that is still a hard challenge to get 17k requests per day and that will be 12 per year, I want a decade of data, so that makes 17k*12*10 requests. Not a small number to deal with. Its not rate limiting but processing that much is quite time consuming. Even parallelism won’t help here.

Anyways…

Will check over the weekend if there is a possibility to pull it off quickly and cleanly.

Quote of the week

“One machine can do the work of fifty ordinary men. No machine can do the work of one extraordinary man. ”

― Elbert Hubbard

This hits home for the week. I had Codex, amp, and Gemini run for me for the flight observatory. But none of them could do a single thing without my instincts and pulling the cords. It was almost as if it can know a lot of stuff but cannot move a penny, only when steered in the right direction by the right person. Like an aeroplane in the hands of a pilot. It cannot do it properly without the steering and right guidance, the right mindset, and experience. Be extraordinary, don’t let AI limit you, let it push you.

Created

• Case Study for Mumbai Airport with ADS-B Historical Data

  • This is half baked, and made an AI-mistake that it took 900 second snapshot when the data was available for 5 seconds. Just because I told you I want the data to be quickly downloaded.

  • Have spent rather a lot of time wasted in this downloading part, I want to analyse the data, but its quite hard to pull it off.

• EPL Dataset Exploration for CS50 SQL Final Project

  • Demonstration Video

  • This was a great side quest for the final project. All done easily with a simple dataset. Downloaded the dataset of matches of EPL seasons from 1994 till 2025 and showed them to a sqlite databsae.

  • I got a great deal of info, who won the most epl titles in the past 3 decades. Which team scored most goals, and the silver lining of all was the validation query that I ran “which team never lost a game in a season”, gave the perfect answer. Arsenal, the invincibles.

• Tags and a separate RSS Feed on the blog

  • I wanted a single layer for my other places like s3g to pull my content from a reliable and non-blocking place. So this was much needed.

  • Now you can fetch tag wise, post type wise and even all content feeds.

Read

1. A eulogy for Vim

   • A really heartfelt post. I was not expecting AI, but yes, it matters.

   • Bram Moolenaar, an absolute legend, the creator of Vim. He died a couple of years back. The author of this post just wrote his heart out. Every humanly possible connection is relatable (not for me for Bram, but yes can see that). He was kind, he supported poor people in Uganda.

   • This post just gives me a hope that people can live still. AI will probably eat the world, but people like him will cease to be eaten up. And that is liberating, that is moving thought. People need to be aware what is happening in the world.

   • Vim Classic, a tribute to Bram, really great stuff. Will support if I can with any power.

2. DatabaseMaxing with Preston Thrope

   • This is a great guide. Inspirational and eye opener. I wanted to do it, year back. But still procrastinating. I had written about learning SQL, but now in the phase of building low level stuff like interpreter and compilers.

   • I am thinking of building a markdown parser like pydantic for the web. I know that is wired but it can open up a lot of possibilities.

3. Yes, I am bored of reading and listening about AI too

   • I hear you. Managers and literally everyone are somehow shoehorned into software development for asking how many tokens they are using.

   • Do they even know what a token is? Do they know wha a database engine and storage is, their difference? The difference between a row and a column storage nuances? Why? Why not ask those, instead of obsessing over the metrics. Its a wired and frustrating time to be a developer in one perspective.

   • Yeah! I know its all great and wonderful time to be a developer as well. But the bad parts just suck out the joy of it. The analogy of discussing what tool to use and how is just infuriating. Just show what you build and how, not with what and nerd sniping the markdown file specifications.

4. So where are all the ai apps?

   • Good questions and modest exploration. I don’t think the increase in productivity of building of software has any relation to packages being developed. Since the first trend that we would see is that people will make software for themselves. Solo software or personal software as many have called it, apps are just that.

   • Yes, that comparison of packages being shipped is a good point but not fully true and valid. Its kind of very early (oh, we are 3 years into AI now). But people are still figuring it out.

   • We say what just happened with LiteLLM on PyPI. That might be a nail in a coffin for shipping pacakges. Or even installing them.

5. To live in a world without AI

   • We have a flurry of posts like these. Good to see that. It gives hope that there are people who think what I fear is happening.

   • The thought from waylonwalker is truly a good callout. I hope we stay on the thinking loops rather than sitback and let clankers do the rug pull.

   • I am really startled that, this AI dystopia has already arrived and if it goes any further the world looks like a bad world from a fictional book. Phew! Turn of a year or two and bamm into a dystopia of robots.

6. Every Kubernetes concept has a story

   • A beautiful post. I was confused midway, should I laugh or learn?

   • I don’t think using a database in a container is a good idea in any situation. Unless you have a session-only requirement.

   • The pains of one thing are actually a concept in Kubernetes that is a good lesson to learn and learn each concept with a situation in mind.

7. Markdown ate the world

   • This is really intriguing, I though markdown was old than docx. But its not it was there after 2004, that is quite a recent thing. Just over two decades. And it has changed the way people write.

   • I think majority of the non-tech people still use docx and whatever doc format is for Microsoft Word. Its just works you know. And if it doesn’t work it just doesn’t work. Nothing in between. But markdown always works. Anyways, who can convince them. Microsoft has a deep foot in the minds of people.

   • With LLMs, markdown just became a standard now. Everyone knows it, its simple with just enough structure to separate it out from plain text. Just the right balance but dead simple.

8. Why I vibe in Go and not in Python and Rust

   • I can feel this might be written by AI, but still it has a valid point. Go doesn’t gets in the way. Python never does (only in prod), Rust just blocks you.

9. The machines didn’t take your craft

   • A good one, the last paragraph hits home.

     No matter what tools arrive, no matter how powerful they become, they will always remain tools. They won’t replace our reason nor values. You will still choose what is worth building. And as long you reason, nothing essential has been lost.

   • This is a good call to not abandon your craft, just because a faster method exist, never attach yourself to the tools. They will keep on changing.

10. The diminished art of coding

    • Very good. Really good. Its written from heart and humans value this. This is what I call writing. Writing not for attention, writing for connection.

      If you’ve never taken an interest in poetry, or painting, or dance, or whatever, now would be a good time. In an era where the internet is increasingly full of bots pumping their bland bot ideas into everybody’s brains, seeking out distinctly human forms of expression has become vital

    • A good point. In this era, connecting yourself with nature or art as we say is vital. Especially for developers, we need to get out of our heads sometimes, we need to ponder the blank, the boredom.

11. You are not your job

    • Gold. Just sheer empathy and gratitude for the person writing.

      • “You evaluate someone's warmth first to gauge their intent before ability”

      • “The people who love you don't love you because you're good at your job. They love you because of something else entirely. Maybe it's your humor. Maybe it's that you actually listen. Maybe it's that you remember things about their lives and ask about them. Maybe it's simply that you show up. You're present. You don't extract a conversation and then disappear.”

      • “The harder version is asking yourself: if my job title disappeared tomorrow, would I still be me? Would the people who matter still love me? If the answer is yes, you’re in the right place.

        If the answer is no - if your identity is not cleanly separated from what you do for money - your relationship to yourself may need an update.

        You are not your job. You’re a person first. Your ability to connect, be present, and make people feel understood is what makes you irreplaceable to the people around you, which is the only market that counts.”

    • bangers after bangers. Really, I am framing this post. Its like a healing potion. A compass for life in tough times, such as the current ones.

12. Somethings just take time

    • Yes, some things do take time, and friction is what helps it grow. Rightly said.

    • I like the idea of building communities and trust, that is what I had done without having extensive technical skills at my job, I was there just showed up daily, and … it ends wiredly but not all time ends up the right way. Maybe it was my mistake, but anyways, moving on to a new chapter.

    • I don’t know why everyone wants to churn software, what is the hurry, do you know what you want to build?

13. Ctrl + C in psql gives me the heebi-jeebies

    • This is really interesting. I like the way he calls it heebi-jeebies. It really is.

    • Like the TLS is not there for the cancel request, so your psql connection sends the unencrypted database secret in the wild, and somehow if intercepted by anyone in the same network, it can launch a Denial of Service attack.

    • The Neon Proxy and Elephant shark(the wireshark but for Postgres) have a workaround by noting the secret with the initial connection and when the psql sends it with the plain text the secret it intercepts it and kills the right session. Wired stuff but kind of no choice, that would require a bit of a refactor on the protocol.

Watched

• Concurrency Patterns in Golang

  • This was a great video explaining go routines and concurrency patterns. Loved the analogy of gophers as senders and receivers as passing the buckets in channels, the buffered and unbuffered channels as the gophers in between, really well explained.

  • Also, the patterns and concepts for those concurrent go routines and channels were well explained.

• A bad day to use python

  • Wired, how can a pypi credentials be comprimised that too for a founder and that leads to a release without the developers realising it?

  • Very weird, litellm was a great package, I like what they do, I want to make that for Golang. Never able to make one. Thankful that I was not the one with Golang yet.

• OSI and TCP Best Explanation:

  • A very enthusiastic and clear explanation of the IP Model.

  • The difference is really explained really elegantly. The confusion and the separation of layers is pitched right way, removing all the why so questions.

  • Highly recommend watching it to get a good grasp on the fundamentals of the Network Model.

• Composer 2 and Cursor Drama on Kimi

  • Damm! This is a bit sad, but not bad. Like Kimi is happy with the parternship, Cursor played around the limitation. Which is not a good thing, and can lead to other major closed source labs to follow this trend which would make open source models like a free fruit to grab.

  • Not a good spirit, but cannot say anything about it if both of them are happy.

• Kafka and RabbitMQ differences and uses

  • A good explanation of Kafka and RabbitMQ, the difference was there I didn’t knew it. They both are message brokers but one is a smart and other is a storage place with other operations to do things around it.

  • I still don’t know which one to use when, its time to really use them

Learnt

• Downloading a dataset individually from raw files is harder than downloading with S3 or R2 files with RClone

• Always match concurrency to DB capacity.

• Use buffered channels to spawn goroutines.

• Early blocking reduces wasted memory and load on DB.

• Usage of Keep-Alive header in HTTP Requests

  • TCP Connections Are Expensive. Every HTTP request over TCP (or HTTPS over TLS) starts with a 3-way handshake, SYN where client asks to open a connection, then SYN-ACK where the server acknowledges, and then ACK where the client confirms.If HTTPS, we also have a TLS handshake, exchanging keys, certificates, negotiating ciphers. Each handshake takes milliseconds, adds CPU cycles, and adds network round-trips.

  • When we add the HTTP header Connection: keep-alive the TCP connection stays open for multiple requests between client and server.

    Backend (or proxy) maintains a connection pool, reusing the same socket.

    Reducing the Latency (no repeated handshake) and the CPU overhead.

Interesting Tidbits

• https://flighty.com/airports

  • A great UI for viewing map and the traffic.

• https://github.com/ssrajadh/sentrysearch

  • Interesting idea for semantic search based on embeddings.

  • Gemini can understand the semantics of a video from its embedding model and we can search parts of videos. This is really cool, and this is where AI expands into real world use cases.

• https://translate.kagi.com/

  • This is hillarious. It translates sentence to Linkedin like voice. Amazing, whoever made it.

Tech News

• LiteLLM package compromised

  • Phew! That is a wild gun. PyPI attacks have been quite a lot in the past few years. There needs to be some layer of security with it. The ecosystem needs to improve.

• Bye Bye Sora, did anyone use it?

  • Yeah! Like who actually used it? I just saw bunch of ghibli trends and generic slop on linkedin and twooter. That’s it nothing useful.

  • Disney ended the deal, which means either it was not great, or it was too expensive for them. The former is more likely.

For more news, follow the Hackernewsletter (#787th edition), and for software development/coding articles, join daily.dev.

That’s it from this week, it was quite exploratory week. Learnt and built a lot of stuff without actually shipping, but that is where real learning happens I think. Lets ship full guns blazing next week.

Happy Coding :)

Leave a comment

Subscribe now

Share

It was another heavy week, a dark week. Losing grip of mind. Hopes fading. But not everything is doom and gloom; hope can be found in the darkest of times, if one finds the courage to light a candle. I crave reading books, I want to get back into it and build a writing routine again.

I am thinking of doing livestreams this weekend. Let’s see how that unfolds.

Quote of the week

“Failure meant a stripping away of the inessential. I stopped pretending to myself that I was anything other than what I was.”

― J.K. Rowling, Very Good Lives: The Fringe Benefits of Failure and the Importance of Imagination

Really hitting the quote. When you actually fail or life fails you, you begin the quest to find your true self. There is a path to redemption. The only path carved by ego that leads to good and intended places. The self-identity and self-respect when it gets hurt, it bites the soul to the core, and some energy has to come to the mind to steer it in the right direction, here the energy might be negative, but the intention is always nobel. The only time you use your ego in a good place, rest all places its a waste and a waste of time and energy.

This is the time I will find myself, I think. It might come with despair and a lot of burning inside, but I can say I might not regret it later.

Created

• S3G API for querying my posts with SQLite hosted on Vercel Cloud function

  • I wanted to build a pipeline like this

    curl -s https://sqlite-static-site-generator.vercel.app/api/query \
      -d '{"sql":"SELECT title FROM posts WHERE section = ? LIMIT 5","args":["newsletter"]}' \
      -H 'Content-Type: application/json' \
      | jq -r '.rows | map(.[0]) | join(", ")' \
      | xargs -I {} ~/.local/bin/llm -m gemini  "Summarize: {}"

  • Pretty cool if you ask me. Like I can get insights on my blog without getting archives and downloading stuff from different sources.

  • Just get the json > process the rows > pass it to llm and ask things.

  • Will be sharing about it over the weekend or next week.

• Flight Observatory

  • This was a random hit idea. To build a dashboard about Flight and air traffic around Mumbai. I wanted analytics for past decade or so. I wanted to know what has changed, what impact COVID had, what impact the new NMIA airport has, what the flight patterns are over the days, months and seasons, also how the monsoon affects the flights.

  • That is a big ambition, and I failed to do extensive research. I just spawned a few chatgpt threads and tried to get out of a good plan and hurried to get codex. It spilled the beans. It is live, but not what I wanted. I haven’t got metrics and the right insights in place. I am impressed by the amp than the codex, to be honest.

  • Maybe the free lower models have failed me, but whatever.

• MicroGPT fork that learns per-document representations and, for each generated token, estimates which document contributed most to that prediction using per-doc probabilities. More on that next week.

Read

• What is Agentic Engineering

  • Its kind of wired how the shift has focused from agentic to all sorts of things, in 2025 it was buzzing like crazy. Right now it has gotten better but still is bit hyped and not established. Everyone has a huge divide of opinions.

  • But the article did a good job of breaking it down and making it clear and to the point.

  • It addresses the main problem in engineers these days “Now that we have software that can write working code, what is there left for us humans to do?” Yeah, we have a lot of stuff still to do. Writing code was like taking a stroll. Honestly I think writing code by hands gives me a time to slow down my thoughts and think it through. Now that with agents, I have hundreds of thoughts spiralling and zig-zagging my brain that I can hardly do the deep work that was once possible and is possible by hand coding. I am not saying to write code by hand, but it is a therapy to a developers crazy buzzing mind.

  • This is true

    we need a term to describe unreviewed, prototype-quality LLM-generated code that distinguishes it from code that the author has brought up to a production ready standard

  • We desparately need a term for it, otherwise, people are confused with agentic coding and vibe sloping. And this article neatly does that.

• US Job Market visualizer by Andrej Karapathy

  • Wow! What a use case to explore. He is a genius. Every week, doing some novel stuff.

  • I spent 5 minutes finding where the word software was. Rolling eyes (I know)

  • It was bewildering to see that it won’t grow as much as people are thinking. Its not a sought career maybe now-a-days. The frustration of change is going to bite the newcomers like crazy.

• Agents cannot replace thinking, research and direction

  • Well noted. Really, I experienced it myself when agentically coding the flight dashboard app. I knew what I wanted roughly, but the limitations were really bugging. I threw everything in the sink that was available, and it did a good job. So yes, research is vital and so is thinking a bit.

  • Without a direction that you decide, it will definitely hold a wrong direction.

Watched

• Requestly Schema design

  • It was interesting one to see the modeling of API client. I was intruiged by the introduction, like why did he start at the variable interpolation. Is that the crux of the product. But yes the flow was good and made all click nicely.

  • However, I think there was more to it, I actually didn’t knew what LLD is. However I thought it might be a little more deeper than the schema. That was good to hear. This level of design is I think a good bridge between hands on coding and system design.

• Stripe logs differently: Canonical logging

  • Totally relatable. We used (still do, *) to do the same thing for the internal service api to log all the metrics in a single log. Whenever we have an event or a block of code or part of module triggered or respond back, we add it to that dictionary and send that full dictionary to the post request log.

  • This was great to see that we built a thing that stripe also uses.

• Its so over: Mitchel Hashimotto take on AI Agents and the grunt work by Primeagen

  • Yeah! I can see this happening. The moat for developers is to know the details and what can be right and tasteful. Experience is going to pay dividends in the AI era.

  • Its not the experience you think, its about how much you care about the details, if you already went deep into thinking and researching each api you interacted with, AI is going to make you hell of a person. If you are not, hmm, might be some skill issue. But its kind of bad that it will create a huge gap in developers.

• Are designers cooked? helll no! Stitch by Google

  • That looks slick. A good interface I must say, away from the sidebar code editor and things like firestudio and what not. This feels really good.

Learnt

• Clickhouse basic setup and analytical queries

  • I learnt clichouse a tid-bit to get a taste of what column-based databases actually feel like. There was no difference in the usage; the use case, however, is the point.

  • Its all plain SQL but the time and the way we think about data is totally different. Its meant for read-heavy and analytical operations.

• There is no free forever analytical db like duck db or clickhouse on the cloud

  • Yes, there is no. I was thinking of hosting the 5 minute cronjob with a clickhouse database that crunches the data for months or years. But nope.

  • Cockroach DB, and others have free-trial and not free-forever tiers. Such a sad state.

• API to get last 5 minute live airspace data

  • https://opensky-network.org/api/states/all

  • This was a cool find. I almost got carried away and built an API without researching what were its limitations.

  • I wanted a dataset for past decade or atleast half a decade worth of flight data. I wanted to build an analytical dashboard for BOM airport. But couldn’t find a complete one, so landed on this somehow.

Tech News

• Astral to join OpenAI

  • What is going on here? AI labs buying runtime companies and products? Like what? Don’t they have enough coding power to build it.

  • Ah! They buy the mindset and the people not the product, is it?

  • Not sure what Anthropic will be cooking with Bun and now OpenAI with Astral, but it sounds exciting to be honest.

• OpenAI releases GPT 5.4 mini and nano

  • Hmm! I must say, OpenAI is the only lab that consistently drops models weekly. Anthropic is busy with suing people, I think.

For more news, follow the Hackernewsletter (#787th edition), and for software development/coding articles, join daily.dev.

Its all suing each other, and layoff things. Quite a negative environment to be in. I wonder to escape in a cave and live a peaceful life, sigh! But you can beat the negativity if you don’t see it. That is delusional but you need to be if you want to go big.

Happy Coding :)

It was a lonely week. Trying harder then ever to survive. No work. Yeah! Hint hint. I am not at all bored and sad, but trying to gush out all the projects that I can at my disposal. I have a agent running all the time, I switch accounts if one gets exhausted with free trial, Codex(perfect yet lower models are a bit dumb), AMP (blazingly fast) and Gemini CLI (I hate it).

I am building new stuff, just trying out to be there. Its hard than ever to be a software developer. I never thought this will be it? Like I wasted my 5-7 years in learning the craft? I don’t know if it was all worth it. Seems its hanging with the mercy of the capabilities of an LLM call. Really sad times to be in.

Still hopeful, there are a lot of problems to solve yet! Humans are on the cusp of a make-or-break. World War 3 is looking imminent.

Quote of the week

“To overcome the fear of failure, have the goal worth failing for”

— Paul J. Hutsey

A goal worth failing for. Hit hard, doesn’t it? if you fail, your life was worth the effort. It might have given inspiration, a sense of hope for others to rise along with you. You might have failed, but have woken some dead people along the way, that is worth it. Its not sacrifice, its doing your part right. Its your responsibility to work for your own goals, no one is going to walk it for you. Yes, you are alone walking, but as we walk, the path appears, people come as if from thin air. Just take the step. Failure is not fatal, success is not final.

Created

• SQLite Static Site Generator (S3G): s3g.meetgor.com

  • A SQLite Shell for querying my blog posts.

  • It also has html hydration from URL > write the query > view the results > .share > paste the link and view the full post in html

  • Just a side project that I wanted to build just becuase SQLite is everywhere and can be used to do anything. Also its’s a good way to expose my content. I realise I have more than 800 posts.

• Crossposter Agent: github.com/mr-destructive/crossposter-agent

  • Just extending the fun project I had 3 years ago for the crossposter shell script.

  • Converted that to a golang service and added AI to it.

• Revamping meetgor.com site

  • Added search and yearly archives

  • Pulling link-blog from the past year and newsletters

  • Fixed some issues with tagging and ui

Read

1. AI slop terrifies me

   1. Good enough? That’s a myth, in AI world, everyone is blinded by it. There is a 3 phase thing, > let me try if ai is good > oh this is GOOD, even GREAT, but good enough > oh damm, we need a developer to fix it.

   2. This is true to vibe coding, this is vibe coding, if you generate slop, you are a vibe coder by that I mean you don’t care about the thing that solves your problem you just care that the problem was solved. I don’t want or even like a bit about being that kind of person. If I was, I won’t be here writing this post.

2. Can LLMs be computers

   1. A very, very cool post. I am intruiged that a transformer can itself run the code, or atleast become the computer, that is so powerful.

   2. The phrase is quite substantial: “The key difference is that tool use is opaque: the model hands off control and receives a black-box answer. In-model execution is transparent: every intermediate step appears in the trace, and the model never leaves its own decoding loop.” Yeah, that could really help the transformers become more capable and resilient to tool errors.

3. I am not consulting an LLM

   1. Banger.

      experience is where intellect actually gets trained

   2. Yes, LLMs don’t lie but they lie well enough, good enough.

   3. This traces back to “are you willing to explore things” or “I just want the fucking work to be done”. If you are the former, LLMs might brain rot you and you’ll be back to normal. If you are a later, you will be a LLM maximalist. There might come a time, when both of them meet.

4. Why I put my whole life into a single database

   1. Mind Boggling. Like how did he manage to capture all of that? I can’t think of a tool for each activity that I do, that would be bizzare.

   2. Maybe Apple tracks all of that? Safely of course but still that is a lot of data.

   3. I am intruigued, there is a lot of stuff to see here. I am drawing some inspriation from it.

5. Why everyone is talking about filesystems

   1. Interesting very interesting. The analogy of at protocol of social systems with agents context is really intriguing.

   2. Actually this is really correct. The filesystem is really the crux why software has evolved if you see databases and CRUD apps. Everything is a file.

   3. Skills are actually just md files, that I already said. How fascinating that just a few markdown files can change the way agents can behave. Someone just mentioned Moltbook, yeah!

   4. Also, this is relatable to me because this week I built S3G, which is basically taking all my posts and dumping them to a single json or SQLite database, which is a single file. Wow!

6. Software ate the world, now AI is eating the software

   1. Damm! This hits hard. Maybe true. But the point is change. Software is changing, the crux remains the same, to solve problems, AI address the how part, and not the why still. So, the tools with which you did your problem solving will evolve put the problems might still remain with increasing complexity.

   2. This quote is correct here fits nicely: “Computers will still be, as Steve Jobs once put it a “bicycle for the mind” - just with completely new tools that don’y need to be explicitly programmed”

   3. Ok, the last sentence had me in awe. Programming computers was never supposed to be a human job anyway, really? What about the 50 years of toil? Like you can’t call it hobby, its a field with research and lives poured in it. Cannot be wasted due to a pesky auto-complete bot.

7. What do coders do after AI?

   1. Bang my head, please. This post just made me smile.

      More than anything else, what I hope people can remember is that all of the great things that people love about technology weren't created by the money guys, or the bosses who make HR decisions — they were created by the people who actually build things.

   2. Yes, there are hard parts, but it’s put in beautiful words and respectfully so. It is a reality, and the faster we accept, the better we position ourselves in the future.

   3. The two groups that this post described, the 9to5 and the hobbyist, is intriguing, I don’t know where I fall. I am drawn by money, but I am here because of my interest in the first place. If money was not a reward, will I have been here, yes hell yes.

   4. Layoffs, that part hurts a bit. Change is the only constant, but the writer mentioned that this change is the worst he has expeirenced. What about the new joiners, new grads for whom this was the only change they witnessed. I feel for them. Its tough. But its not doom and gloom, humans have found ways in toughest of times, and the humanity shall find that kindling hope again.

Watched

• Pewdiepie’s model beat ChatGPT?

  • Whoa! That was amazing one. Bro just rickrolled Sam.

  • True signs of a grug brain developer.

  • The way he just tried and tried after failure is so inspiring, people still need motivation? This is the best video I have watched so far.

• Zero Project ideas?

  • I don’t know, that is very broad topic. I have a lot of ideas, but none of which crystalise into an actual product maybe

  • Maybe now I need to think that deeper. But building is something I loved, but LLMs seemed to have snatched that too from us.

Learnt

• In Websocket Protocol, the server doesn’t mask payload whereas the client has to.

  • Because to avoid cache-hits on the payload if the message gets interpreted by the middle servers, it might never reach the actual server.

  • Really great insight, I am using ChatGPT daily to learn and study mode for certain topics that I want to dive deep, it gives a little paragraph that I need to answer, the perfect way to engage.

• We can embed data in a URL to hydrate a webpage

  • I just built S3G and it can basically render pages just using the data in the URL as a state (title and content). This is the same thing used by excalidraw to encode data with sharable links.

  • This way we don’t have to query server for that data. Really a good piece of architecture design to remember. If the data is not sensitive you can save a round trip for the client to the server.

  • Here’s a demo: s3 sample sharable url

Tech News

• Meta hires the duo behind Moltbook

  • Wired, why that was just a slop show right?

  • Maybe the mindset of building was what got them at that place, that is not deniable but anyone could have built that, if one really needed it.

• Atlassian to cut 10% workforce in pivot to AI

  • This is coming, AI is taking and chipping in piece by piece and we don’t raise any bells yet.

• Perplexity opens waitlist for Personal Computer

  • Ok, this is just like claude cowork? Perplexity was ahead for a while, but I don’t think they have any moat anymore.

• Anthropic releases Claude with the capability to generate visuals and interactive elements

  • This is coming to every lab now. They are making it easier to learn with LLMs.

  • I don’t know why they want to shoe-horn LLMs in everything.

For more news, follow the Hackernewsletter (#786th edition), and for software development/coding articles, join daily.dev.

That’s it from a hard slogging weak, it feels like shouting in the void. Bad time passes is the slogan I hear not from anywhere, just within. Anyways, hope you are coping with AI too, if not you should.

Happy Coding :)

It was a life-changing week, for good or bad, not sure. But something is sure, I have made decisions that might change the coming months. Sometimes I get in a rut, especially since AIs have made it harder to think outside the box. Now is the right time to change gears.

I am excited. I have a wide open life, I feel like a free bird. I can flap my wings and choose the direction. I cannot be more privileged. I am taking time to learn things more as usual, but now, I think I will learn the old school way, hands-on, not agentic things and all. I need to level up.

Software is a field that demands constant adaptation, but this change is quite hard, every change is hard. We need to find a way through it, not around it. Let’s see what we have in the bag.

Quote of the week

“If you(your life) are about to take a leap, you’ll need to take 2 steps back, you do it willingly or life pulls you back”

— Meet

If you have been experiencing a setback, don’t fear or get scared or heartbroken. Get excited, life is preparing you for a leap of faith. Trust the timing of the universe, trust the direction, if you did everything right till now, consider it life’s intertwining to uplifit you, get to your destined place. I am not talking philosophically, it is science. If you want to jump, you’ll have to take two steps back, reflect and then act. You can’t jump from the current position, even if you can, you can’t move much further.

Things keep changing and it won’t be long before everything gets clear and you’d realize it was all working to make it fit together, they were puzzle pieces trying to fit.

Read

1. You don’t have to do it if you don’t want to

   1. Man that is banger of a post, I just read 10% of it and was smiling and was in peace. content.

   2. There is someone who understands this pain of working with AI.

   3. I want to read this article, its too long, but I want to feel it. Weekends are for that.

2. Antidote

   1. The article has lived to its name. It truly has and is an antidote. Just read it. Read it I said.

   2. Just do things like you used to do before AI, AI is just a tool, it is not necessary to shove it all the places and situations. Be candid, be original, make stuff just because you can.

3. MicroGPT

   1. This is beautiful! I used custom list of names of Pokemon, Places and LLM Model names.

   2. This is so cool, I still don’t get it, I want to read the code, and build something different from it. So much to do, so little time.

4. You can use newline characters in URLs

   1. This is wired, I never thought about it, does that really work? I can see this being used for Base64 encoded images.

   2. Worth knowing. Better presentable HTML.

5. Is traditional software engineering dead - Naval Ravikant

   1. Nope, this is true.

      But software engineers still have two massive advantages on you. First, they think in code, so they actually know what’s going on underneath. And all abstractions are leaky. So when you have a computer programming for you—when you have Claude Code or equivalent programming for you—it’s going to make mistakes.

      • It gives me hope.

      • Not sure because, he is not a software developer, so cannot really fathom how he can commet those all things, but good points.

6. 747 and Coding Agents

   1. Really good comparison. I feel like a pilot than an engineer now. Really good.

      Yes, the learning is becoming the most least focused thing.

   2. Finally people realise that it is happening. The change is too big of a deal, and its just shaking the grounds of software like crazy.

7. Stop using Pickle

   1. Pickle and waste?

   2. Ok I can understand the pain here. The pickle file is not readable except for that python program.

   3. By using some other standard format, it can help in reusing of data

Watched

• Peter Steinberger with OpenAI and OpenClaw

  • He just knew a lot of software, so he can do it, I am not there yet, I need to write and read a lot more code to be there.

  • This is fine, I think, as long as we can understand what the code is getting generated, I would feel nice to just let AIs do it.

• Software Engineering is dead now

  • Ah! This is bad. I didn’t realise it till wednesday. Things just hit like truck. Really sad for so many people, atleast they have 6 months, though the times are tricky, the hiring might be wired place. Its not the same, people have confusion on what actual software would mean in 1 year of time. I can hardly think what I will work with in the next month.

• Did Kellogs do the math right?

  • Nerd stuff. Really cool to see it.

  • Really, spheres can cover more surface area for filings than donut shapes, really intriguing.

Learnt

• TCP Protocol.

  • Yeah! I didn’t knew how it worked internally. I learnt that it:

    • Opens a port for listening to clients

    • Client connects to the port with 3 way handshake

      • SYN (share the client sequence number)

      • SYN-ACK (server recieves teh sequence number and sends its own sequence number)

      • ACK (client sends the server sequence number +1, indicating it recieved the servers sequence number)

    • Read from the client body

    • Write to the client

  • Its basically a protocol to communicate, it doesn’t define what to communicate with, the client can send anything. There are protocols built on top of it to do specific format request and responses like HTTP, Redis, SMTP and dozens of others.

• Tail -f

  • We can listen to file and it can automatically show the last 10 lines if there were any updates. Just wow, perfect for log analysis or monitoring

Tech News

• Google releases Gemini 3.1 Flash lite

  • This is a bold claim, that a cheaper and faster model is better than 2.5 Flash. Flash was a one of a kind shattering all records model when it launched last year. I am really looking to test it out on few things. If available on free tiers.

• OpenAI releases GPT 5.4

  • Damm! OpenAI just keeps banging version bumps one after other. Every month they have something to release. Just wild.

• Apple launches Macbook Neo

  • Well! Now people would be seen with office and college laptops as Macbook Neo I guess, cheaper and premium look. Do they really care if video editing doesn’t work, probably not.

For more news, follow the Hackernewsletter (#785th edition), and for software development/coding articles, join daily.dev.

That’s it from this week. It was a tough one to digest, but now there is no turning back, oh I have already two steps back, time for a leap of faith and change. Cannot be more excited for the coming week, for what life has in the pack.

Happy Coding :)

It was a lazy week, I’ll put it straight. I don’t know what I should be doing or not doing. LLMs are crazy. I can get lost in them without realizing how much time has been spend (also cost). Same thing happened with books this week, I almost read 700 pages this week. I don’t know how, I just was lost in the books. I finished the harry potter series. I was struggling with the half blood prince on 25% last week, but I powered through it and completed both the books and here I am.

This is not a book rating newsletter but I say one thing. Reading books can change your perspective and not restricted to any one field. I think I saw a lot of inspiration for fighting against the dark arts or LLMs. No, not fighting against LLMs but sort of understanding when to let it rip through itself and when to strike. It was like Horcrux and Hallow. Not quite but something. No I am not trying to mash up anything I read to LLMs, it was just natural. I’ll leave it for another article.

But the point is, its getting lost. The art of code, its already is. We can’t write, the hard work is automated, just like that in a puff. There were people around me who always told me “Code is the last thing to care about”. But I was of the other opinion. Because writing code gave me clarity, it gave me direction, it was a validation of where I am going. But now its not quite like it. You can’t enjoy without feeling. I am an idiot here, stubborn talking about it. Next week I might be euphoric about using LLMs. No point in blabbering and hating it.

Quote of the week

“Do not pity the dead, Harry. Pity the living, and, above all those who live without love.”

― J.K. Rowling, Harry Potter and the Deathly Hallows

A touch of gold that quote is. Summarizes the full series in beautiful words. Just brilliantly done. Changes the perspective of the mind, the intention and the affection. I was literally crying after reading that, I was numb after reading that quote and didn’t read for minutes. That line just moved me.

Pity for the living who live without love. What a hitting stone. Pity the LLMs, pity the people who can’t fathom the care for art. I am just dropping that quote here. I just have no words to describe what I feel. Its beyond senses. Beyond the power of words, because to feel the words you need to experience pain and life, for life is sometimes pain.

Read

1. Writing code is cheap now

“The biggest challenge in adopting agentic engineering practices is getting comfortable with the consequences of the fact that writing code is cheap now.”

1. • Banger, yes can agree to that. Turning that code to production grade is hell of a task. It seems so flawless that its tempting to just push it straight. But tests my god tests are like verataserum for those ai slop.

“Delivering new code has dropped in price to almost free... but delivering good code remains significantly more expensive than that.”

1. • Yes totally.

“These best practices are still being figured out across our industry. I’m still figuring them out myself.”

1. • Honestly he is the best one to put this way. Everyone thinks they know how it works, but having that probabilistic factor is very rough especially for thinking of code in a new way.

2. Hoard things you know how to do

   • This is an interesting idea, I can think that those all things that we can do could be a test on how good or bad the models are getting. Right now there are a few things that I can do and feels a bit awkward to do, so LLMs are a good point to tuck in.

   • Also the labour job of writing code is away now, no denying that. It was in the ChatGPT phase too. But now it can do at scale with the full context.

3. Google API Keys weren’t Secrets but then Gemini arrived

   • This is funny and scary at the same time. Funny because how carelessly the single API Key exposed the full Google access, scary because how can such thing at Google scale happen.

   • Kind of wired how they handled these. It felt a bit rushed and then never looked at. They wanted AI to be in the hands of everyone and everything. So I think that might be a decision somewhere that for AI studio it should be that but they mistakenly made it for all products accessible from the API Key.

4. This time is different

   • It is always different right? In tech, no change is the same, that’s why its called change right? But people think every change as revolution, and this is like that which hits like a truck after the hype has faded out.

5. Google street View in 2026

   • The indian region looks full of street views, shows how conjusted and cramped everything in India is. Peace is a luxury in India, though people have a unique way of finding peace in chaos.

6. We build a app to read books with LLMs - Merrilin

   • This looks good, I want to build something like this. Recently I have been devouring books like crazy and have found myself typing 100s of queries to google (not GPT) to understand the plot deeper and certain quotes that I didn’t understand.

   • Might not be a good thing to add while reading a book, but as a companion to talk to is a good direction.

Watched

• Anthropic is lying to us

  • I don’t want to start another week of dunking on Anthropic but this doesn’t seem to end.

  • They think using their APIs is against rights, but scraping internet and training claude is not? Well they should get more of these now, let them taste their own medicine.

  • I am starting to get a hatred for them now. Can’t bear them.

• We used to be gamers

  • This was a fun one. Really enjoyed the banter. I was also a novice gamer with my friends in teens. I used to play Clash of Clans, Clash Royale, Brawl Stars, Minecraft. Those were the times.

  • Now it feels like a dread and creep, almost like wastage of life, not time even. I know, I know I am not a productive-rambling person. But it just, I don’t like playing games anymore, there are other things for me to enjoy and sip my soul in.

Learnt

• Its quite a problem to now say yourself a software developer.

  • The right term would be a “good software developer”. Because software has become a commodity. Models can write code and build software, but “good” is a relative term.

  • The taste, the experience, the curiosity and the grit. These qualities are not in LLMs (yet). But there are others which humans can’t compete like speed, intelligence and availability.

  • I don’t know who would trade of which things for humans but it looks a good trade of if someone just wants to get the foot on the market. The bar has not lowered though, the steps have. The gate is still tall to climb.

Tech News

• Nano Banana 2

  • This looks another bump in image generation models. I think image generation has plateaued. Except for Sora and Nano Banana, we haven’t seen any growth in the past year.

For more news, follow the Hackernewsletter (#784th edition), and for software development/coding articles, join daily.dev.

That’s it from this week. Hoping to get something out in the next week with code. Weekend is all packed with plans. A bit humanesque-friend gathering after almost a year, I am tired of talking with LLMs. See ya next week.

Till then keep churning tokens!

Happy Coding :)

A pretty heavy week. Things getting cooked up for developers. Shift in how to develop features. It feels a bit awkward that management people are teaching developers how to code now. Wired times. The craft is officially dead, long live code writing by hand.

I am sad, but not that sad. It feels like something valuable is being snatched from my hands, but something more powerful is given back, but I don’t know how to process it. Just like when our parents have mobile phone in their hands for the first time. Or we have rubiks cube in our hands. Its quite a powerful tool, LLMs. For that I don’t have a brainpower to process what it spits out, the steering and orchestrating are the skills that developer like me lack like shit and this is exactly what we are put in.

I am ready for a change. I don’t resist it, its just a wired times where all are learning how to play with them and nobody knows what and how to deal with them. They think they know if they have gotten one or two projects right, but the models are evolving at a lightning fast speed. Keeping up with all of that is a chaos.

Quote of the week

“A bird sitting on a tree is never afraid of the branch breaking, because its trust is not on the branch but on its own wings.” ― Charlie Wardle, Understanding & Building Confidence

Great quote. And rightly so for todays developer facing the dilemma of AI threat. Do you trust your fundamental skills or AI’s next token prediction? That quote just liberated me from that question. And honestly it came naturally, I accidentally stumbled on bing search due to a link opened on my browser to it, and I clicked on home, but it redirected me to the quote of the day and it was that. Wild! But yes, we need to trust the learning that we have gained so far, the hardships, the errors and bugs we have solved so far. I sound like cringe right? But for a developer battling and grappling with these questions is in need of these words to help him understand he is not alone. Everyone is figuring things out, everyone is one that kind of branch which can collapse anytime, might not, but you have wings you know how to code, so don’t you worry. You can fly if things break you, a bird never things aeroplanes can threaten them. Do they?

Created

• GCP Log Explorer TUI

  • This was a long due project. Finally made the courage to throw it with full force. Its not complete but is way better than what I would have imagined, it is functional and has a ton of options. Maybe I need to think again of the UI elements and decisions, but its a good start.

  • I built it with amp for the initial scafold, its quite fast it churned the full project within 10-20 minutes. Then it was broken badly, nothing was working. I handed over to codex with the tmux capture test idea and it just ripped the project apart for 1 hour and gave that project that we have right now. 0 lines of code touched by me. Mind boggling.

• ClickUp TUI

  • ClickUp oh dear! The time I might have waited for it to load, would have been the equal amount of time writing in the clickup space. The UI is horrifyingly slow.

  • I though TUI might help, but I think I am getting proven wrong by themselves, their API seems to be slow and the UI bloat is baggage burden.

Read

• Why I am not worried about AI job loss

  • Humans will still be the bottleneck. I kind of agree to that. Because in the end it is humans who will perceive the tasks, they can’t have super-intelligence if none of the people who work with it are even intelligent.

  • Guys, learning is going to be quite critical. The past couple of years have changed the way we perceive learning. I think we are getting into a trap of outsourcing the thinking and eventually learning to LLMs, which is looking a bad direction, and the turn needs to be as steep as possible to get back on track.

“GPT-3 has been out for six years; GPT-4 for three; and none of that has happened. Even in the outsourced customer service sector, the lowest-hanging fruit on the automation tree, we’re just not yet seeing mass layoffs due to AI. I’ll be frank in telling you that this has been a huge surprise to me. (And to others.) There is change, but it is gradual; it looks more like standard technological diffusion than a tsunami of replacement. And we should think seriously about why this has been the case.”

• • This quoted paragraph gives me hope to continue learning more.

“people have responded by spending much more time coding than they used to, because the latent demand for software is so enormous.1“

• • This I must say is true again.

“If we don’t need jobs, we’ll still invent them”

• • Yeah! That is the spirit, that is the mindset people need to inculcate, and not panic or get lost in the existential dread. I am saying this to myself, because written words have power over vague mind conversations.

  • Its going to be fine. Humans will live or die, either ways, it doesn’t even have a 0.000000001 % or 10^-100000000000000000 effect on the universe.

• Building TUIs in easier now

  • Banger! Banger of a post. It just blew my mind, when I asked it to test with tmux.

  • Yeah! I created 2 TUIs on that day. One is complete 90%) functional and here is the link. The other one is janky, because the UI is too.

  • I loved this article. It gave me a good advice to test tuis since it can understand text, tmux has options to capture text from sessions, which just open a wide variety of programmable automation and testing.

• Peter Steinberger to join OpenAI

  • This is huge. It also is showing the rift building up between Anthropic and OpenAI. The one taking the advantage of the mistakes of the other. And OpenAI I must say has not placed a foot wrong in 2026. Anthropic on the other hand has ruined itself with a few already.

  • I take my words back for now, Anthropic was Gryffindor but it choose to be evil and should be in Slytherin. OpenAI I am not sure it is brave but so is Google. For now, OpenAI is Gryffindor for me. Brave and Generous, expensive yes but better from the competition.

• Desensitized to AI Hype until tried Opus 4.5

  • Yes,very well put. The gap in everything is wide. The shift of perceiving software is changed 180 degrees. There is wide gap of what code is generated and what is shipped, the knowledge of developers, the usage of models, the landscape of product, its all widening like crazy.

Watched

• Opus and Codex Models

  • Another banger! Sort of proving my experience too.

  • I was working with amp (it uses claude models in free tier with rate limits) for creating TUI for Clickup, I asked it a feature for opening the task when supplied with a link when opening the TUI. Like clickuptui --link clickuptask-link. It was not able to load or understand the things. It added the feature but was not working. I asked it to fix it, it was not working. It just removed the feature! Like what? It just removes the problem out of the way rather than untangling it.

  • Then I switched to codex and it solved the problem, slow yes but it did it.

  • Amazing how each of these types of LLMs are evolving.

• Anthropic is a Cult

  • More and more things are getting verified it seems. Anthropic is just on a brag mode. It thinks it is a superior or a pure-blood kind of race. Really they are wired about how they perceive intelligence.

  • I am annoyed by them now. They have good models, but the vibes are not feeling good.

• Dario Amodei - Dwarkesh Patel Podcast

  • Cringe as hell! I feel a greed and haste in earning profit and not humanity in the sight. He gave a example of curing diseases, but has he thought what is the other side of this mess? They are sort of up to something which is not quite clear.

  • The idea of “country of genuises in a datacenter“ is quite ambitious and good, but is the curing of disease the only task? Is it only to replace talented humans? Replace art with slop? I don’t like that thinking of automating the intelligence part. It just gives too much knowledge without our brains having the speed and capability to handle.

• The real reason Anthropic built a Compiler

  • This week is dunking on Anthropic and I will do it with heart. I also thought it was “from scratch”. Well, there are quite a lot of astericks forgotten by them.

  • Prime is right on the take away being, we now have agents that can coordinate for a task which can be weeks long, but Anthropic is suggesting something that causes panic and existential threat. It sound good on their words but if you just think it becomes melodrama once you see the details.

Enough Anthropic dunking, we need some other lab to step in and be a worthy crown for coding models. Deepseek V4 around the corner? Can it beat Claude for coding? Let’s see!

Learnt

• Agentic usage on development

  • I understood on a deeper level to work with Agents.

  • Ask LLMs to ask question to you. That sounds wired but think for a moment. When you used to code by hand (back in the days you know), we developers used to ask questions and get answers by compiling the code, seeing the output, and iterating. Same is the case with LLMs, they need a opening to ask questions, they code, they run code, but where is their feedback? Letting them ask you questions gives them the context in a much better and concise way. It also forces you to think about the problem you are solving.

• We can use tmux capture session to get the screen captured for debugging tuis

  • This was a great leaning from the blog I read and the playing around with LLMs, it was fun and exciting to see how to intuition is still vital in dealing with LLMs for coding.

Tech News

• Google releases Gemini 3.1 Pro

  • This is quite quick. It is better from 3 Pro but still not significant. Its also I think the first time Google has release 0.1 increase in the model, it was either 0.0 or a 0.5. Something different.

• Qwen 3.5

  • It is quite neck in neck with Opus 4.5 and GPT 5.2 in terms of benchmarks.

  • Open source models are really keeping up with the closed sourced labs, I must admit there might come a time when that would have exceeded the closed source lab capability.

• Claude Sonnet 4.6

  • Meh! I don’t care about Anthropic anymore. Its just annoying now to think about them after understanding their vision.

• Sam Altman and Dario refused to hold hands in AI Sumit

  • This not stopping guys. That is clearly a rift, a rivalry, a gryffindor and slytherin rivalry. One taking the other. Tables turn pretty quickly.

  • That was embarrasing and quite evident that dario is some wiredo. lets call him that wiredo.

That’s it from this Anthropic dunk and AI revolution week. I hope I come out strong in the coming months and weeks.

Happy Coding :)

Week #81

Its kind of a wired position to be a developer. People are talking about replacing or atleast questioning developers now. I am in awe, how a field that was the most harder to work in, suddenly gets bambolzed with little ghosts with token predicition machines. Just think about it, the industry that spent millions and billions in humans to solve problems is getting wiped out in months, if not weeks. I am scared!

Maybe it might just be the last year I be a software developer, might have to find other field to work in. Kind of feels tragic. People are making me even more scared. Do you know what to do as a software developer? A person who has spent half a decade in learning how to code, realises that the same thing can be generated in seconds. If there was an artist, who spent a decade in creating a painting, yet with AI does it take away his soul? No right? I also think coding is like the same, yet is the most least bothered thing. People don’t give a shit of how you solve the problem. And that hurts as a developer. It really does.

Quote of the week

Curiosity and Enthusiasm to learn cannot be diminished with any force of nature

I don’t know if this quote is true anymore. But I think, I will still cling to it. As again I don’t want to repeat it but being a developer the old way is no longer feasible or required or even admired. It is a new era of generation of code, you can’t start having attachment towards chilsed code, it can be replaced and generated in matter of seconds, with the right mind and intentions.

That’s why for those right mindset and intentions, a flickering fire of curiosity with a kindling touch of enthusiasm is a sustatinable approach. It might not be, for I don’t know what the future holds, I can just believe.

Read

1. Stop using Icons in data tables

   1. This makes sense. Text makes it easy to view without the cognitive load and stuff. Really nice on the eyes too.

2. Why is the sky blue

   1. Nothing tech, but a good article. I just like to read and don’t mind learning something out of the blue. And why actually is sky blue.

   2. I think more content should be like this, interdisciplinary and broad topics. People need to now think about overlapping things.

3. I am happier writing code by hand

   1. For the past 6 months, I was, but after a few couple of months, it feels like people are no longer in sympathy with that feeling.

   2. There is no care for code, it was a art, well now it might be the lost art. I know it is hard for developers to accept it, but change is something we have in our blood. But man this is not change, it is just erasing the need to write code by hand. People are just managing this little agents instead of files now.

   3. Everything is a prompt they write, a little nudge is what they see instead of a little read. I don’t know where this is going, but it can’t be reverted, that is for sure, the drug is real and it can’t just stop.

4. We mourn our craft

   1. Damm, every post I read is this. We might be the last generation who remember writing coded by hand. Wow! We are that last era.

   2. I don’t know to be proud of it or be scared. Its hard to see anything as a developer now. Am I just a prompt writer? Just a system person thinking about the problem or what even is the need to make products?

5. Heroku is finally officially dead

   1. We saw in 2022 what happened and now this is the final nail in the coffin.

   2. PaaS king that stayed for half a decade now, is almost dead. Flyio and Railway are the new kings.

Watched

• Sam Altman and Theo on the future of code

  • Its unncertain, but yes the learning problem in LLMs is qutie nasty.

  • It doesn’t have a constant memory like humans, but it has a good brain, which might be mometary, but exceeds the capacity of humans. Maybe that is a wired statement, but it lacks something humans have, yet has something that humans don’t.

• Creator of Clawd on the Pragmatic Engineer Podcast

  • This is wild, he was always a nerd, a curious person. He has built a ton of things before many of his things have gone viral right? Maybe its not true for him.

  • Its kind of crazy how he has just made so many fame out of building something really valuable, but then it feels almost like anyone could have made it, a problem first mind comes into picture.

• Agentic Coding has a problem

  • Yes, this is a problem I face, but I am not calling it a problem. I am not a fullstack guy yet, or atleast I don’t shift projects that radpidly.

  • I am using the same things I used to use, tmux/zellij and normal editor in my workflow. I love agent in the cli, its great, but now I realise it is a token hungry thing, you don’t see on the screen how much junk or slop it generates behind the scene, when suddenly your cursor prompt says, “Quota limit reached”. Yeah I have been there.

• GLM 5 is a great model

  • This looks a leap in a good direction. Atleast we have a amazing open wieght model. Yes its not self-hostable, but we can use it to some very cheap price.

Learnt

I don’t know if I should learn anything anymore.

Yet, I have curiosity, but sometimes, the curiosity flame is wavered by a gust of doubt and guilt.

I still have a belief, I have a purpose. For that I will cease to die.

Tech News

• Go 1.26 released

  • Does it matter? Yes it might.

• GLM 5 Released

  • Really good to see open source models having such a neck-in-neck competition with the closed source labs. Worth voouching for.

For more news, follow the Hackernewsletter (#782nd edition), and for software development/coding articles, join daily.dev.

It was a exciting week. I went to co-working space which felt like ages. Created something meaningful with ai-assisted coding, and it worked charms. It helped get out of a dry slump and make better impact with code. Its me, in my mood, when I have the fire, nothing survives in front of me. Just that the fire is lost in the mundane dance of life, it throws you in valleys you never knew existed or had planned for.

Back to a grind of coding with LLMs. This week a lock in mode to ship 3 projects. Let’s see and what I can come up with. Maybe I’ll write an article about how my experience has been so far.

Lots of possibilities and exciting times ahead. Get out of your limited thinking, and just sit with your mind, negotiate and build good habits.

Quote of the week

Ram is the heart, Krishna is the mind

In the battle of kalyug, the heart is like ram and the mind is like Krishna. This is really beautifully put. You have the best of the previous two generations fighting against each other. One on the right path, the other on the right mindset. Both are important, none is completely outruled. You need peaceful mind, for that you need a pure heart like Ram, but people will take advantage of it, without a right mind, you cannot be at peace. You’ll have to survive the duel between the both, don’t get caught in the trap of helplessness, decide when to act and when to restrain. The person who can be both at the right time, is the winner here.

Created

• The Daily Token Newspaper like blog

  • I used meta-ai llm free, and some vibe coded slop to make it. I felt good. Used Gemini and Amp to make it.

  • The meta-ai just stopped working the moment I created this page. I have to patch the library to make it work now.

• Gemini Reverse Engineered Client

  • Gemini is the only now that offers free limited chat without bot detection. At least as of 6th February 2026. It can change the moment I post something, such is the speed of shipping of people now.

• Youtube Cookie session Exporter

  • This is something I felt good building, atleast watching. I made it so that, I can copy a session from my un-authenticated profiles of browser and transfer it anywhere I want to browse feeds and take a different spin on it.

  • I discovered that the expiry of these tokens is 6 months.

Read

1. I miss thinking hard

   • Boy! That hits like truck. Absolutely relatable. We can share that part of our mind, it lives rent free on my mind. How to balance the builder and the thinker part of my brain.

   • I also had feelings like actually typing the code gave me the time to think about it, but now the time between prolonged thinking is just squishing like thin line. Its getting too much building and no thinking or taking a step back.

   • Maybe that is how we will move forward, but it doesn’t look sustainable. Developers will burn out and eventually give into AI slop. But here we are learning to deal with them at the moment, and it seems we need to find a way around and through them and not out of them.

2. The most important thing while working with LLMs

   • This makes sense. Just like children, when you say something, they will go wild and try to interpret what you actually meant to do, they will circumvene around the instruction bt won’t quite follow your exact instruction. Its not controlling, its programming the model.

   • The process of making it faster by breaking the steps into parallel is quite interesting. Not sure if everything can be done that way, we would spend so much time in thinking about how to break the problem which is half of the solving part.

3. My AI adoption journey by Mitchel Hashimoto

   • That was a banger of a post. True and Honest.

   • Chatbot interface was a good gateway drug to ai-assisted coding, but its not quite a good one, move on to agents with cli or tool access.

   • Check if it can do what you can do, that is a good advice and pratical one. It gives you the taste of what can work and what cannot. I have done it myself for one of scripts to get metrics from logs, all of which I could do in half an hour, but with agents and right context, I can now do it in minutes. But that took some time to understand what to give it, and what to not.

   • Keep an agent running, think about what you could be doing but can delegate. This is quite a good advice given how smart they can get given the right context and tools.

   • Do the work, till the agent does its. Don’t delegate and chill. Forming skills is something still valuable as a human.

• That is damn point

Well, you’re trading off: not forming skills for the tasks you’re delegating to the agent while continuing to form skills naturally in the tasks you continue to work on manually.

• • Use agents.md, skills or whatever the harness can use best. Its a ongoing process but don’t get caught up for long in old ways.

  • Always have an agent running (maybe just one). That is a bold advice. And this leaves me with disagreement for quite a subtle reason. It can feel like I am missing out on something if an agent is not running, seems like wasting precious time in making something, which is not a bad feeling to have, but can ruin the day.

Watched

• Theo in Opus 4.5 with Skills: The best model for frontend design is...

  • That was cool. I didn’t knew Gemini was that good without skills at frontend design. Need to actually try it out.

  • Also its quite comical that a single md file can steer a atrocious model like opus 4.5 at design into a marvelous tasteful designer.

• Theo on Codex: OpenAI just dropped their Cursor killer

  • This is perfect for me, i have 5+ tabs of agents running on the terminal with cursor-agent cli. I crash the system very hard. This is really smart thinking.

  • I thought of making terminal interface for the vibe coding a particular prompt into multiple models as worktrees. But this is taking into a different league.

• ThePrimeagent on Moltbook failing

  • Yeah! That was just a open database. Humans manipulated to make it look like llms did it. What a shame to be a human. Why do they need to make it act like agents did it, if they can’t then simply say so, and even if they can’t its not of any use.

• Theo on Moltbook situation

  • He was clearly hyped about it. It felt like sci-fi to me when I watched it. But the next day we say the crash, the reveal of the project.

Learnt

• Skills/just markdown files can make LLMs better

  • I saw this with Clawd bot or moltbot or openclaw or whatever you want to call it. The molthub or moltbook is based on just a single markdown file.

  • This is just mind-boggling like reading a passage from scifi book. The transition to scifi from normal mundane human life was just a matter of a text file. Damm! That lasted for a day, but it was a good run.

• Agentically using LLMs is the way to go

  • I created something that increased the metrics by 15% and I was really happy, keeping the cost at bay and increasing the numbers which was like very hairy problem. That was with agentic coding.

  • Love to see that how things will turn out in the future.

• Cursor doesn’t count tokens used in the agent-cli in the analytics page

  • I use agent-cli all the day, and I was in the bottom 10 of the organisation chart of cusor usage. I was like what the heck?

  • Cursor cannot configure its analytics right, Its not just me you know. Only two people would get this line.

• Sharepoint deletes links after 90 days of subscription cancellation

  • Yes, this is subtle but very ridiculous from Microsoft. It makes sense, because then whats the point of the subscription if you can revive your files anytime, after years and months.

Tech News

• OpenAI releases Codex agent management GUI

  • This is something solid and I am looking for something similar to this from open source community.

• Anthropic drops Claude 4.6

  • Hhh? We were expecting Sonnet 5, but what a surprise drop from Anthropic. Did it not succeed at the evals or was it worse from the intended score?

• OpenAI drops GPT 5.3

  • Surprise. How can OpenAI have 0.3 versions. I have never seen 2.3, 3.3, 4.3 but a 5.3 very well, very well. Looks like the plateau of models is inevitable

• Mistral releases Voxtral 2 transcribe

  • Mistral struggling as always from the beginning.

For more news, follow the Hackernewsletter (#781st edition), and for software development/coding articles, join daily.dev.

That’s it from this long, energizing week. Talking to people helps. Its not what we would do by automating, its a question is what can be automated and what should not be. AI is changing things fast and quick. Developers are at the forefront of it. Can’t be more blessed with a perfect time.

Leave a comment

It was a tiring lazy week, honestly. I am not not-excited. I am excited, but it feels pointless. Like swimming in a pool where people are crowded. I don’t know what I am saying but building things has never been easier and the floor is open but the barrier to reach user is still the same.

This weekend I would like to complete some unfinished things vibe coded projects and put it out. I already have completed one, there is one in progress, the last one will complete and consolidate it as my thing and put in the web. I love these times.

Quote of the week

“The bigger the goal, the bigger the hurdles, you’ll be surrounded sure, but keep the fire flickering, you might die giving a hope for someone behind”

I think this is true. I have bigger dreams, ambitions, maybe bigger responsibilities. But that doesn’t give me an excuse to give up. Time and time again, I am shattered, broken and bamboozled with problems coming out from nowhere. I still keep on the hunt, I can still see the prey, I need to move one step at a time.

I think its arguably evident that the bigger the calling (call me religious or philosophical, but I believe it) the bigger the forces to stop you from doing it. It can be a deadly weapon, every weapon is two sided. Blessing can become a curse in a flick of thought. We need to understand the core idea of the decision and then track back to our intentions.

Created (Vibe Coded/AI-Assisted Coding)

• CMS Finally working with Netlify Cloud Functions + Hugo

  • I finally spent time to fix my link blog setup. I made sure the weekend it was smooth, frictionless as possible, and exportable in markdown.

  • The post I add on CMS gets saved to DB, the cronjob on github action picks up the saved post in the past 6 hours and adds a file in the repo.

  • This builds a fresh site every six hours at https://notes-meetgor-com.vercel.app/links/ which I’ll port to my own site at meetgor.com

  • It was fun, I did it like a ralph loop with Amp Code. It was fast. I use Amp code just because its fast. Not necessarily smarter, but it has some edge on Gemini CLI for sure.

I would be vibing this weekend for making some products end to end. I have been inspired by some people at my org, leading from the front. I love to be in such a place and am blessed to be here. Would like to show and prove that I too want to see success.

Read

1. One Year with Kagi · Blog · Eli Perkins

   • This is nice, a good insight actually.

     • Don’t outsource thinking by reading the AI overview from LLMs

     • Human curated lists are often good and better for your brain than AI slop

     • Search engines already have biases in them, by putting AI they are adding a new dimension to the biasness

   • Kagi is something I have installed on my phone but didn’t quite use it. I think I am getting lazy and am taken away in the habit of reading the AI overview which is one click away.

   • Need to rewire it back to good old days with Kagi it seems. But there is also a counter point of searching and getting information effectively maybe that is partially true if we are outsourcing our analytical thinking in searching and skimming by reading less articles.

2. Some notes on starting to use Django

   • I wonder if django is like unintentionally made for LLMs

   • It has everything suitable for context

     • great docs

     • 2 decades of stack overflow questions

     • robust and explicit

     • not exceptionally magical

   • It just makes sense, its like a mechanical part of a system, others might hide complexity or maybe too verbose

   • But django just hits the harness the right I think. I have read couple of articles on this and I think it makes sense.

   • Though for now I favor golang instead of django why? type system. You can add pydantic or mypy in django but out of the box support is where I am inclined towards for now.

3. After two years of vibecoding, I’m back to writing by hand

   • True, maybe. It feels it kind of trashes the way through the solution rather than path finding to a solution

   • The image is so well presented, the idea hits home.

   • Though I think only certain people are able to get value out of it, its a skill issue which eventually everyone will cope with in the end.

   • Not sure how well good or bad it is, it seems to be fading out now.

4. AI coding agents for enterprises | Mistral AI

   • This is finally something I have been waiting for.

   • An agent free, to run. Remote agent.

   • Claude Code, Codex, Opencode, require some paid tier.

   • This looks like I can finally use one from my phone.

   • Jules surely is there but its so buggy and just halts for no reason, not reliable enough. Might make my own agent.

5. Agentic Table Merging | Tensorlake

   • This looks really interesting.

   • Very close to the problems that I am solving. People are trying hard on agents and this I thought was far fetched, but maybe not. Agents are the way.

   • We need to find the way through agents and not out of it.

6. AGENTS.md outperforms skills in vercel’s agent evals

   • Passive context (AGENTS.md) currently outperforms active retrieval (skills)

   • Skills are still useful for vertical, action-specific workflows

   • I think I can say that LLMs are bad at reliably picking tools, skills, or docs. If the information is needed, make it always present rather than calling it separately. The best results for this eval came from removing choices, and ambiguity

7. Inside OpenAI’s in-house data agent

   • Instead of requiring analysts to manually explore dozens of tables or write intricate SQL, the agent lets them ask plain-English questions and get high-quality, correct data insights in minutes instead of days.

8. Litestream writable VFS

   • The big thing I learned is that SQLite can now pretend your database lives locally while secretly pulling just the tiny pieces it needs from object storage, on demand.

   • That means apps can start instantly, even with huge databases, and only hydrate the data they want which is wild if you’re used to slow restores or heavy disks. Wow

   • Instead of copying data to compute before you can do anything, you let compute skim data lazily and write back carefully. It’s a clever trick, bending old constraints without breaking SQLite’s mental model, Flyio cooks wired and quite intruiging stuff.

Watched

1. Laracast: I’m Done

   • Laracast cut off 40% of the workforce, sad again AI in the hunt, after tailwind this is sad.

   • They are producing more content but the way he thinks about code is changing.

   • The important thing that hits me is Agentic Coding doesn’t drain mind. Is this true? I don’t think so. I feel like when I used to program, I thought about what to do, then plan it out, and actually writing the code would give me a buffer, a mental buffer to calm my mind from the actual cognition, it triggered a different part of my brain. But right now with agentic coding, the phase is too short, and it doens’t trigger different parts of the brain, I have to review code which I am learning to, but it feels like I am getting too much load on the thinking part without actually taking a detox from it. It might be just me but this I need to change.

   • Maybe this was the next iteration of programming, no one knows, but a good thing to see people admit.

2. Which Programming language for AI

   • I was wondering the same, but I read something like LLMs are good at typed languages. Its not quite true though it seems. Rust and C++ should be shining here, if that was the case.

   • It actually depends on the ecosystem and the core principles of the language and not just the technical features of the language.

• DHH: Why AI isn’t writing my code yet!

  • Oh Yeah! Oh no! I thought it was LLMs don’t feel good to code, but the answer was awkward. He is previleged (earned not luck) with the luxury to be in a position where he can code with hand chisels, rest of us have to slog with LLMs to make our day job.

  • Its a harsh reality, the writing code part is becoming a hobby rather than a job I think. The vibe-slop cleaner is more of a job now-a-days.

  • I love problem solving but sometimes AI is too fast for me to walk, I can’t run all the time, I am not saying I am lazy, but writing was the perfect thing to spend my time thinking and tinkering, LLMs seems to take away that time and replace it with hollowness and existential threat.

Learnt

• Elixir is a better programming language for LLMs to understand and reason with

  • Its quite easy to read and understand, everything is immutable makes it hard to shot yourself on the foot

  • C# is also a good contender, not just the language design, but more so due to community and documentation.

  • I am coming to realize that having a great language is not a technical feat, but a mutual agreement between usage and depth. Like LLMs struggle with Rust too due to too verbose and cult like culture. Golang is not quite impressive as Elixir due to lack of proper documentation and tooling. Python is a sandbag, filled with good and bad examples on the internet.

• Coding and Reviewing are like Writing and Reading, you need to do both

  • Coding is a mechanical task, you can get away with it and feel good

  • Reviewing on the other hand is not a straightforward approach, you might feel good, you might feel dejected, you may learn something, you may waste time, its quite a bag.

  • You can’t keep doing one and expect things to change, you need to balance both.

  • Just like reading and writing, you need to feel inspired after writing, or get your thoughts out after having a giant reading spree. Both are needed, don’t become a elitist in either of them, you can have taste in one of them, but don’t overlook the other.

Tech News

• Moonshot AI releases Kimi K2.5

  • The model seems good. I love the way it reads. I hate when GPTsque models give list after list. This model just reads like a breeze. But oh they have improved the coding side of it. Nice side quest.

• Google DeepMind releases Genie

  • This is quite a good thing from Google. They are starting to explore world models. Which could be a incremental steps after LLMs

• GrokAI releases the API for Imagine

  • This is a controversial feature, I don’t like the direction of Grok on that feature, it’s quite unchained and unguarded. Opposite of Anthropic, Its like a Slytherin vs Gryffindor. I am not liking when I say Anthropic is like Gryffindor but after Grok is doing I need to.

For more news, follow the Hackernewsletter (#780th edition), and for software development/coding articles, join daily.dev.

That’s it from this week, it was a slow week, had a long weekend, but looking forward for a new week and a new month. A perfect month.

Happy Coding :)

Leave a comment

Share

Some weeks are no for hoarding knowledge, I realize this after writing this edition. And this week is that one, the one that you need to put breaks and not reflect that’s done, but see the road ahead, not how far we have come, but where to go next. What to do next, why build something, why spent 2 years on something. Question, think, understand and let it settle. Everyone is doing that. Some aren’t privileged as I am, I am grateful for that.

I am going into the long 3 day weekend with a lot of projects to build, would be disappointing to not have at least one hack from the shed on Tuesday.

Quote of the week

“I carve things of wood because things made by effort are more real than things made by wishing.”
― Katherine Arden, The Girl in the Tower

Oh yes! This hits. I read the book last week and completed the trilogy yesterday. The quote hits home. Why? Just compare the hand crafted, effort rich written code with AI generated code. I guess I don’t have to speak anything more. The rest is on your own right? You can express you feelings about one or the other. It just is similar, but maybe incorrect. Code is a means to an end, just like some tools in our lives are. If I am a writer, laptop or a book is a means to an end, I can’t obsess one over other. I would use either whichever is at my disposal.

Effort and wishing would come in to play when you are doing it for your or others pleasure and to express something which wishing or words couldn’t.

Read

1. The recurring dream of replacing developers

   1. what a flashback, everything makes sense.

   2. Programming is not mechanical

   3. yet people tried hard to make solve for it as it was mechanical

   4. COBOL made syntax readable. CASE tools eliminated typing. Visual tools eliminated syntax. AI can now generate entire functions from descriptions

   5. Each advancement addressed a real friction point. Yet the fundamental challenge persists because it’s not mechanical. It’s intellectual. Software development is thinking made tangible

   6. Just sheer facts those two. We need to find a way around using AI not away from it.

2. The joy of being a competent beginner

   1. This is really well put. Very relatable. We all have started with some quick competence at something in the beginning and then ignored or abandon after some familiarity of it.

   2. This exactly lists why we do that, and the reason is that going beyond that beginner competence is a steep learning curve, initially you are fast but then quickly hit a wall.

   3. Most of them give up, the ones that stick, are the ones that somewhat develop a mastery or sort of craftsmanship in the art.

3. Welcome to Gas Town

   1. Agents as code generation orchestrators. This is quite a ambitious thing. I haven’t read the full post. But I can see where it goes.

   2. I haven’t either been in situations with 10s of these agents ripping in the background. There are reasons for them and some of them I am trying to overcome.

      1. I don’t have that many ideas honestly, this is flawed in my opinion, my biggest weakness maybe.

      2. I don’t have claude code as I don’t have much to spend on AI. I am limited by free options and some work related subscriptions. I am bogged down by the clumsy free models.

      3. The habit of abandoning a project after the setup is too evasive now. The earlier habit has just got more notorious with AI.

   3. That all said, I love this idea, the next step in AI is agents orchestration. Maybe I am behind it, but I had a few ideas, not exactly this but some level of parallel agents running. Not orchestrating. Maybe that idea is speaking more than ever. I thought someone solved it, but nope. Need to roll up the sleeves.

4. The bet on juniors just got better

   1. This is fair. The bet on junior aka me was that I will take the ownership. And this previous year I did. They might have gotten the payback but not quite like AI. The thing that AI might miss is reliability. Not availability.

   2. If something goes wrong, I can wake up and roll in. But if some non-informed developer or AI does it, there it could get into a different rabbit hole.

   3. I think the more quickly you can show your eagerness to solve problem, actual user problems the better the bet payoff would be. Its not rocket science but is easier said than done.

5. The challenges of soft delete

   1. Nice read. I had experienced it in my first internship. This problem of dead objects. Especially if you are using Django and Postgres. It looked easy to add a field of soft deletion. But the resulting queries could create bottlenecks.

   2. Since then I haven’t quite gotten the chance to explore this, this article showed me the different ways to implement the soft deletion.

6. The toil of blog art

   1. Yes, the art of expressing some concept is valuable. AI might have eased it, but true human-eque art is impossible to replicate. The chef’s kiss is what the author is trying to meld in the post about.

7. Writing first tooling second

   1. True, gold. This is to the point and another way of saying, “Show, don’t tell”, so “Write, don’t setup”

   2. The blog is merely one possible organising principle, not a requirement.

   3. I started the same way. I picked up hashnode, then moved to jekyll and github pages, then used python via markata (waylon walker’s ssg), an ssg that someone else wrote and I loved it, it was what I needed, the control of what goes in and out and also it was easy to see what was happening so that I can change and remove what I wanted.

   4. I finally now am rolling my own SSG in Golang and a CMS system. I am not very consistent in sticking to one cms, I have built 3-4 versions of them. But the thing is I still post consistently. Not long form posts, but these reflective posts and short bursts of thinking.

   5. Write your brain out first, then eventually it will outgrow to your needs, the system will be formed not shoved in. Like earlier I just used to write long form how-to-guides or tutorials, then I started to write reflections weekly, then link posts tils and suddenly I had 10 types of posts. Article, Tutorials, TILs, Thoughts, Link-blog, Newsletter, Notes, and what not.

8. Why talking to LLMs have improved my thinking

   1. I like the take here. The one aspect of it only.

Writing has always done this for me. What is different is the speed

1. • That sentence just changed my perspective on LLMs. I was skeptical but now, since they have gotten the powers of thinking, tool calling, I think they are good at talking out ideas and forming maps of different features.

2. Nested code fences in Markdown

   1. This is clever. Never knew this.

   2. I think this clears the rule of when to escape the backticks and the fenced code block within one. Really nice to know this. Helps in writing as well as developing a SSG.

Watched

• Amp Inc. Raising Agents: Episode 9

  • Writing code by hand is over?

    • There will be things where you will have to write the code, but like assembly

    • it can just do things, like give me a cake

      • It doesn’t need the recipe, or hand holding of each task, it can just do it

      • Taste it even and then check if its burned or not, it has a taste or evaluation thing as well

      • It can think about things

    • You need to make the codebase ready for agentic ready

      • It needs harness for testing, good documentation, edge cases, actual problem it solves

• Claude Cowork: AGI is here, hheh?

  • I loved the video. It showed the flaws and the possibilities of this tool. I think its a step in the AGI, but good or bad, the people will decide.

  • The edit button on twitter, that had me rolling out loud. It was a human-esque reply though. “I can see the edit button therefore I am logged in as ABC person” True. Good thinking Claude. Hope you continue in a limited set of thinking.

• Claude Cowork

  • Again, I was tempted to learn how different people perceive this tool. I never watched her videos. But this video came to me at random and I thought of watching it, it was fun.

  • It also showed a good starting point and a legit use cases for people to curse themselves a little less with such tools.

  • Developers can do it with writing scripts but laymen can’t oooohhh. This tool should just do that.

• We need to talk about Ralp loops

  • It kind of is hillarious of how this works! I am not able to wrap my head around it. Like why and how

  • What kind of ... Writing it immediately hit me. I do the same thing that Ralph loop does with AMP code free tier.

  • The context size is limited, so I have to be wary of the limit, I keep the summary of the thread once the limit is reached and continue a new thread. Wow. Writing actually makes things visible and find the hidden patterns. Gold!

  • Watching this video now makes sense. It just a loop for agent to start from where it left off without bloating the context. Superb.

• Its time to change your database - from Supabase to Convex

• • Oh! The convex database now makes me in awe. It never clicked and all of a sudden it rings bells and whistles. The schema changes the database, that is wild.

  • I can see myself using it for my favorite language. GO!

  • Its statically typed, so I can catch bugs before hitting them on the database.

Learnt

• We are in a time of both awe and suspicion

  • I am in awe that I have a tool that can do a lot of things, on the other hand I am kind of having existential crisis on what my job might ask me?

  • Like I don’t have anything to add in learning section, I am just figuring things out to learn. I guess some weeks you can’t force yourself to learn technical things, its ok and even necessary to let go of not hoarding knowledge or information (like I learn how to do x in y, etc)

  • Some weeks are for converting the knowledge and connecting them to form insights and developing a intuition for the ultimate wisdom. This is that week for me it seems.

  • I honestly say, I didn’t read a lot of code, atleast not something out of which I knew. I fixed bugs, yes, wrote code, maybe, but generated code, hell yes.

Tech News

• ChatGPT users to have targeted Ads

  • Its happening. The inevitable.

• Anthropic release Claude’s New Constitution

  • I want to read it, will surely read it. But I am tempted to watch theos video first. Will detail read it over the weekend. Looks something is spicy brewing in the AI mind.

• Winapp from Windows for making windows application

  • Kiind of wild and wired. Who makes them like that? I thought it was like a Microsoft Copilot slop for AI Agent. Thank gosh it was not that.

For more news, follow the Hackernewsletter (#779th edition), and for software development/coding articles, join daily.dev.

That’s it from this long slow slog week. The weekend in India is long. So will be taking some time to invest in learning and building some slop from LLMs. Maybe will build a better intuition for working with LLMs, I know its not a one day process, but I have spent some months in it and continue to do so. See you next week!

Happy Coding :)

It was a harsh week. Not the roughest yet tiny exhilarating with new hopes. It happens, to get the best of me, things will come towards me with force. I am welcome to those challenges and hurdles.

I am in awe and inspired to build things, I went off track last week. The previous week I was pumped with two side project in a day, yet slumped then. This week would like to build that momentum back.

Quote of the week

“You need to control the mind. Thoughts will tempt, but you decide to act or let it go”

I must practise it heavily. Working remotely is tiring in a different way. Sometimes you just have no one to talk to, no one to understand your mindset. I get caught in overthinking and contemplating and procrastination. I need to divert myself to other things, its not like I don’t work. I just cannot resist AI to delegate the work and let me read more articles and watch tutorials and videos. It hard to control the mind, but I think its more important in being aware of it.

Read

• Don’t Fall into the AI hype

  • This is interesting and it comes at the right time

facts are facts, and AI is going to change programming forever It does not matter if this or the other CEO of some unicorn is telling you something that is off putting, or absurd. Programming changed forever, anyway. What is the social solution, then? Innovation can’t be taken back after all. I believe we should vote for governments that recognize what is happening, and are willing to support those who will remain jobless. And, the more people get fired, the more political pressure there will be to vote for those who will guarantee a certain degree of protection. But I also look forward to the good AI could bring: new progress in science, that could help lower the suffering of the human condition, which is not always happy.

• • All points and counterpoints are well addressed here. Innovation can’t be taken back, that just hits hard. Harsh reality even. AI is in the wild, you can’t avoid it, you’ll have to capture and understand them, just like pokemons. Its hard at first, but never say never. I learnt the hard way, kept giving AI things, tried different things, modes and models, and found the mindset shift. I found what I cared the most, but also a part of me felt taken away.

• A typical PDF document

  • This is cool, I’ve read a lot of these and working at docsumo, makes me want to read about them more.

• Ideas are cheap, Execution is cheaper

  • Oh, my god. This is a bitter truth. Geez.

  • Never thought about it. Really its kind of true now. It just is a quick change.

  • The mindset shift is critical.

• A Letter for those who fired Tech Writers because of AI

  • True empathy is key Liability, everything becomes liability if outsourced.

  • LLMs don’t have the taste, the care feeling of the users, the developers yet.

I read a few other posts, but haven’t gotten a chance to sit on it. Will roll a blog for such link post, vibe coded yes! Ideas are cheaper, executive is cheaper now! Hell yes.

Watch

• Kubernetes Zero to Hero Course: Alta3 Research

  • This, is a masterpiece. I learnt everything. Like atleast touched on everything. Loved it.

  • Want to get into it, need to leverage it and play with it.

  • It clicks to me now, the autoscaling, security, the volume bit wow. Everything makes sense after using them and taking them for granted due to cloud run abstraction.

• 2026 Standup Predictions by Teej, Primeagen, Caesy, Trash

  • Pure entertainment. Great insights.

  • AI coded bug yep, that is happening and people won’t notice until a month, bold one.

• AI Assisted Coding

  • Cool advice. Need to improve on clarity, delegation and orchestration. That is a pillar in system thinking I believe.

Learnt

• Cursor doesn’t count the tabs,code,diff s if not done via their UI

  • Kind of frustrating to see myself down the chart for the usage in organisation. I use it heavily. Yet I am looked as AI skeptic.

  • Frustrating to see their own analytics not getting into account. The acceptance is rubbish, you should not measure acceptance just measure the generation bit. No one will let the code be unused right?

• Kubernetes Fundamentals

  • It has three components, Kube API, scheduler and worker

  • The pods and kubelet and all are confusing yet good explained with hierarchy.

  • The kube api gets the manifest and delegates to the scheduler and other parts to spinup and manage resources.

Tech News

• Anthropic launches Claude Cowork

  • This is like a first step in laying the ground work for unemployment. Yeah really. Look at it. It can do a lot of things. Menial things that got humans paid for.

  • It might create panic in industries. A good product but quite threatening and unbelievable.

  • We saw it coming, but its too soon. We are in the start of 2026 and the wave hasn’t yet subsided.

For more news, follow the Hackernewsletter (#778th edition), and for software development/coding articles, join daily.dev.

Leave a comment

Share Techstructive Weekly

It was a good start to the year, finally doing something that I had struggled to do for the past year or so. AI Assisted Programming. Yeah! That was something I finally somewhat understand, and can do it without feeling a slightest of grudge or emotional drama. It took a while to realize it, but here we are. 2026!

I don’t know how I feel right now. Its quite a good times to be in tech.

Quote of the week

“If you want to know what a man’s like, take a good look at how he treats his inferiors, not his equals.”

— Sirius Black

― J.K. Rowling, Harry Potter and the Goblet of Fire

This is true. I think we should treat people with kindness irrespective of their position. Position is no match for one’s love and care for us. They might be doing with their own purpose and needs but they still show it.

Created

• World Atlas Game

  • This is vibe coded in a day. Gemini CLI and Amp. Just ripped it

  • I read Golang was good to work with AI Agents, thought of building some backend with it and lo behold, it did one shot it almost. For the frontend I choose Vue. Surprisingly its a great UI. I am honestly impressed. I didn’t write a single line of code, let even see.

  • I always wanted to make this, but was very lazy to do all of the meddling with the boilerplatey code, it just did in a few minutes.

• CMS with ssg

  • Yes, this was something I have built twice or thrice, it was another shot. I wanted a blog that can just save to the sqlite db and fetch aas cronjob every 6 hours to build the site with ssg. I just gave it and it did. Its not great, it has still qwirks, but making it better over time.

Read

1. Dialogue between a developer and a kid

   1. This is hilariously funny.

   2. What a real developer is? Who knows languages? No, who knows how to code, No! A developer is someone who sticks to a problem when everyone has given up.

   3. This conversation feels like me and my friend. My friend is the reason I am here today. He knew programming well. I was inspired from him, he gave me advice to learn one programming language, I was boasting about python, C and C++. I feel like a kid here. That was 7 years ago, time flies by.

2. Fear is not advocacy

   1. This is real advice. People are hyping about the next workflow to 100x our productivity. Its ok to be 1x and still push less bugs than 100x and push 1000 bugs.

3. Quick and dirty print debuggin in Go

   1. This is cool, we make logging a mess. For logs we need to have separate scripts to get relevant data. How much chaos it can be.

4. AI Did Not Take Your Agency. You Handed It Over

   1. True. LLMs amplify ambguity.

   2. If LLMs don’t have agency, they don’t choose constraints. Well put.

5. On not using Django

   1. I don’t quite get it. Maybe its true. Django provided a good start but then it was like a lock in.

   2. With LLMs its quite easy to generate the boilerplatey code that django provides out of the box, so that demand is lost?

   3. Its not the only reason django is here right? It has extensions, best python community and even more best documentation.

   4. I think it will be the best framework to build with LLMs in the future if the ecosystem continues to improve

6. 6’7’‘ is not Random

   1. This is so true

In the 1990s, a “middle-class job” was enough to buy a house. Being “6 feet” was enough to be tall.

In the 2020s, the middle has been hollowed out.

To be “wealthy” now requires a crypto-exit or a tech IPO (The Economic 6’7”).

To be “famous” requires global virality (The Social 6’7”).

To be “attractive” requires filters and surgery (The Aesthetic 6’7”).

7. AI should be free software

   1. Yikes, this looks like a good take on LLMs being free and open weight.

   2. If not, the larger AI labs might offer ads into the LLM suggestions. This, just the thought of it makes me wiggle with fear. It might push us in wired directions.

   3. The point of drawing a line of “our goal” vs “model’s goal” becomes hazy and it just doesn’t align with human values.

   4. Its a pretty hard problem to solve if it goes in a bad direction, which it seems to be at the moment.

I just completed reading Harry Potter #4 the Goblet of Fire. It was amazing. A good start to 2026 in reading. Hoping to complete the series in February.

Watched

• Designing Data Intensive Applications: Chapter 1 and 2

  • It was a great overview of the database systems. I like how he explains the p50, p90, and all metrics. It makes sense without getting into too much of details

  • Also the diagram of the OLAP and OLTP databases and how it fits. It made sense.

• AI codes better than me, now what?

  • This is really changing. It can write code, better than me. That’s when I started to use it as a partner that knows a lot of things but gets overwhelmed and like a junior does a lot of things.

  • Guiding it, reviewing it, and also understanding myself what it actually does is co critical.

• Database Internals:Chapter 1

  • The difference of the OLAP and OLTP database is so nice.

  • Also the differnce of column based vs row based database type is clear from this. Makes sense and intuitive as well

  • The Binary tree also makes sense.

• The year I stopped writing code

  • This is interesting and eye opening. It actually gave me the reason to be active while working with LLMs.

  • Reviewing is hard, most developers avoid it, that’s the part you need to be doing, in order to be a better one. That point I had ignored and it has came to haunt me in the year throughout. This new year though, will be different. I have decided to take LLM generated code with a grain of salt.

Learnt

1. How to ship code with Cursor

   1. I explained my euphoric moments in the week where I discovered the debug and ask mode in Cursor. It helps me to understand the problem, learn something. Which agent modes doesn’t let me.

   2. I can pause and let it show me what is happening, I can read and share with it, what I think and have a conversation and not just make change all the time. The switching mode was liberating, I think these models should know when to ask and when to execute.

2. Read csv from Pandas in python needs quoted string for multiple commas

   1. If you have n headers and have n+m commas in the row, pandas’ read csv function will break

   2. Because there is ambiguity in which comma is the header separator and which is the actual text comma.

   3. Use quoted string for the text if it contains comma.

Tech News

• Tailwind CSS is in trouble due to AI: Help save the open source community thrive for its earnest effort.

• OpenAI releases ChatGPT health

For more news, follow the Hackernewsletter (#777th edition), and for software development/coding articles, join daily.dev.

That’s it, a good start to the year, looking forward to a good and great year 2026, will it be slow? Probably not, but the year seemed to start slow. Looks good for now, we already have a lot of things already to unpack from the last year advancement. 2025 was pivotal for anyone in tech, 2026 onwards it looks like a year to build and carry that momentum.

Leave a comment

Share Techstructive Weekly

Thanks for reading Techstructive Weekly! Subscribe for free to receive new posts and support my work.

Oh the middle of end of 2025, and the beginning of 2026. This is a wired post. I let myself back, reflected more. I know it was a fast paced year for software development. People are here to prove it. I just laid it out straight that its never been valuable to be a human, a distinct, natural and earnest human.

I must say I completed a non-technical goal of 2025, I have my novel’s first draft in my hands(in my google drive). Its very rough, I want to revise it, I want to remove the fluff, but it is there in its entirety. 33 Chapters, 85K words. 45 days of writing span across 6 months, I did it. I am pumped to revise and write the next novel. The hunger to write has never been higher for me.

Quote of the week

“Grit is never bad, grit with wrong intent, for wrong purpose is definitely bad”

I had a grit to write a story, I will write it. It was a grit, not a goal. I had goals for it for the past 4 years, yet when I made it a grit, here it is, I am on top of it. If a grit was something to think bad of someone, or look down on someone with grudge, it is a bad thing. A wrong grit is something born out of desires, rage, frustration, anger or even jealously, None of it could be bad per see. But if the intention of those desires, rage, jealously is for ill of someone then it could be bad, not if it is born to uplift yourself from the ground, after having thumped by life.

Sometimes, grit is something life give us, throws at you, you need to learn to handle it. It can’t be thought, it needs to experienced. You have to fail, you have to struggle. But here for me, the struggle, the resistance to write finally bent its knee against my grit (not boasting, nor arrogant). Just fierce grit born out of the desire to be a better person, to give back to a human, to say a kind thank you.

Wrote

• 2025 Year Review

  • Feels good to write this posts. I never miss them! I write less post this year. Just 2 or 3, but I have written a lot of SQL like learning log posts (53 of those).

  • Not to forger I am also writing my thoughts on the things I learn and read here. So 52 articles for each week. Easily making up 100 writing pieces.

I completed Advent of SQL with 15 posts too, here are the remaining, which I completed on the weekend. I learnt a lot, it was a good one, the problems ramping up gradually, then the lore for each post was so good.

• Advent of SQL 2025 Day 15: Confirmation Phrase Dispatches

• Advent of SQL 2025 Day 14: Ski Resort Paths

• Advent of SQL 2025 Day 13: XML Travel Manifests

• Advent of SQL 2025 Day 12: Archive Flight Records

Read

1. The future of software development is software developers

technical practices that can dramatically shrink delivery lead times while improving reliability and reducing the cost of change, with or without “AI”

1. • A banger of a post. This is expressing that AI is just a shift in a toolset, or maybe even a abstraction of the language. We will still have ambiguity that a human needs to understand in order to deliver a software.

   • We can see from his experience, developers were written obsolete from time to time, and each time it was different, more potent than the other, but here we are.

On top of all that, “AI” coding assistants are really nothing like the compilers and code generators of previous cycles. The exact same prompt is very unlikely to produce the exact same computer program. And the code that gets generated is pretty much guaranteed to have issues that a real programmer will need to be able to recognise and address

1. • This is true too, we are just automating and generating crap faster, code is always crap until distilled and refine with each iteration to the needs. We just now have a better or worse iteration cycle, a machine that can spit out code like tirelessly, we need to vet and test it.

2. 2025, The year in LLMs: Simon Willison Weblog

   • Boy! That is a lot! I have been saying “overwhelming” word was not sufficient to describe this tend of LLMs in 2025. This explains the reason

   • We had LLAMA falling, Gemini gripping, OpenAI still on the top yet cornered neck to neck suddenly with Chinese Labs and Anthropic in its own league. We saw the sudden rise and sudden dip in vibe coding, people thought “We can be programmers! We don’t need developers anymore, hehe” to “Damm! Do I need a developer to debug this?”. That was a funny thing to watch (as a developer)

   • The images and 6 second video clip generated by AI are mind boggling, we saw from Sora and Nano Banana what havoc they can wreck if put in untamed hands.

   • Local models are getting good, but the speed of the cloud and advancement over the other side is rocketing. There is also this trend of cli based agents. Claude code just set the trend and let 100s of cli agents rip off in the months to follow. Those are still released by new companies every now and then.

   • Slop, yeah! We had less human slop than we needed AI right?

   • Thanks to Simon Sir for this awesome blog. It finally gives me a relief to read so many thing have happened at a glance

3. Andrej Karapathy’s 2025 LLM Year in Review

   • This was more of a reflection post, of how his mental model has changed and how things are building up. I like it. It was a interesting and highly technical perspective.

   • His opinion of LLMs as Ghost is so liberating, as it actually threatens me from my identity if we compare it with humans. Ghost makes sense, even dismissive it as a slave sort of relation right? Not in a bad way but kind of inferior relation for LLMs with humans.

   • Agent that lives in the terminal is practical, for a developer or a human who understands what they are doing, they know what they want, its just too much menial for them to spend the energy on. I agree.

   • There is a lot of work to be done, developers, don’t strap your belts, hone your hammers, its going to be needed.

   • Also his post:

1. • Could this have been more accurate! Right note to end the year.

   • Vibe coding last year, now this is the trend we are surfing on, this will last decades.

2. Local LLMs are how nerds justify a big computer they don’t need

   • Curiosity gets the better of them. I have a 8GB device, I can barely run a 1B parameter model. I get frustrated but have nothing to complain. I can use ChatGPT in temporory mode, or incognito mode if I don’t want it to attach it to the memory. I don’t see using local models on scale is justifiable just yet.

3. You are not dumb, you just lack the pre-requisites

   • Yeah! I have started to learn SQLite and since 2 years made a Brilliant org streak. I feel good taking on advanced concepts soon.

   • Basic and a good foundation helps you pivot and branch off to wide possibilities.

4. 13 Tips for Writing a technical book

   • A handy little thing to remind myself, this is inevitable for me. I would write one. Not this year probably. But I would surely write one, my gut, my instinct is not false on this.

   • I would this then. Great advice for just being curious.

5. Shipping at inference speed

   • This is a good post to say that we have different ways of using LLMs at this point and nothing is permanent. Every month or weeks, this is changing. Adopting a new workflow is like juggling circus art.

   • Codex is something I haven’t even touched, Claude code too, never. I have used Amp, Gemini CLI, Warp and Cursor the most.

   • I love those, those are cheap or even free, they help me understand what I was about to do wrong. They have never produced anything right 100%. I always needed to understand what was I supposed to do.

   • Is this true “>The important decisions these days are language/ecosystem and dependencies” Maybe but I don’t see that. Its kind of true, but not in a big way. The major things are the flow, the edge cases and the intuition for the problem for it to be ale to understand.

   • This actually surprised me”> Go wasn’t something I gave even the slightest thought even a few months ago, but eventually I played around and found that agents are really great at writing it, and its simple type system makes linting fast.” I want to try it now. I have ton of go projects.

6. The internet is a net negative

We’ve maximized information and accidentally drowned wisdom

1. • Hits home. This is good observation and a perfect critique, not over cynical, nor too loathed. Its just helplessness to avoid the battle of the mind and the heart.

   • The business of the world has forced humans itself into a trap. What an irony we live in, creating a cage for ourselves. Besides slaughtering nature into it too.

All that time, that irreplaceable human attention, fed into machines that convert consciousness into quarterly earnings.

1. • That hurts badly. We are loosing are attention to these machines. We need to get it back. The time, the wisdom and the boring tone to our lives.

The optimist in me is still here. Still hoping.

1. • This is great piece of writing. I love it. Want to write essays like this.

   • Thanks for writing this Kenneth, you have inspired some spark for me.

2. What I learned writing Gleam, after coming from Python

   • Top to down approach. This just shifts from taking the problem and boiling it down to the input and output. Wow! This just made so much sense now.

   • We can define the main API as the function that takes something and returns something. In between the intermediate steps, we can then decide what each component of the result will come from.

   • I need to try hard on learning functional programming this year.

3. Engineering is becoming bee-keeping

   1. I like this comparison quite a lot. Swarming agents is what its happening. And the realisation that code was the thing that doesn’t matter, the thing that matters is did we solve the problem

Honey shows up at the end. That’s what matters.

And bees can sting. Without the right gear and practices, you get hurt. The protective suit, the smoker, the careful movements. In code, that’s patterns, documentation, tests. The guardrails that keep the stings to a minimum.

Working like this is exciting. There’s a playfulness to it. You can try things without committing. You can explore without sunk costs weighing you down. You can work on three features at once because you’re not holding all the context in your head anymore.

1. You cannot not lead

   • This is so subtle, yet perfect. You lead by good or a bad way.

   • You cannot say I was not the leader when you are the only person building and maintaining it. You lead by examples, good or bad.

   • Average sucks you know? You are either good or extremely bad. You cannot not lead.

   • Wow! This post is so perfect, not only fits the manager but also every human, a elder human trying to teach or lead a younger one. The younger one learns from the examples and behavior of the leader.

2. Git Integrations is ten years away

   • This is hillariously funny. I can’t imagine VS Code team coping up with git integrations in 2025

   • I don’t blame them entirely, at least they realize it is missing. With these LLM assisted coding, they decided to ship it finally. We have one instance of AI assisted coding helping VS Code ship faster (after 10 years).

   • Learn Git, true. I alway 100% of the times use the cli. NO aliases, no agents, just CLI commands. git add, commit -m, push, pull, merge, rebase whatever. If I don’t know, I google it, read the ai overview and straight to the keyboards.

Watched

1. CMU Databse System #3 Database Storage: Files, Storage, Tuples

   1. This was a good lecture on the different storage hierarchy of the storage. The top there is the pages, the blocks of memory that database fetches for individual records or tuples. Then there is the blocks of memory on the databse file itself, and the actual disk of storage.

2. CMU Database System #4 Memory Management and Buffer pools

   1. Ok OS is not our friend, we need to manage our memory ourselves. This went wild, I thought managing memory was like shooting yourselves on the foot, but not for DBs.

   2. So we load the database file, from the disk into memory not as full, but chunks called frames, where each page is contained in the buffer pool. Interesting, this is done in the actual ram or the memory not full at once.

   3. So this makes it the different algorithms to decide which frames/pages to keep and evict (remove)

   4. There is a difference in lock and latches, a lock is something that protect the database logical content from other transactions i.e. the data to write or avoid corrupted reading

   5. However a latch is something that helps in preventing the database internals from other operations, its only for an operation not a query. Its like a mutex.

   6. We can’t rely on OS, as OS doesn’t know what are we querying.

   7. There are like half a dozen implementation of replacement caches like LRU, Clock, LFU, LRU-K, ARC, etc.

3. How I parsed billions of rows for every user in 2 seconds

   1. Wow! I like these videos. I learnt a lot too. It was passive knowledge true. But I came to know that these things are at least possible.

   2. Clickhouse as a database, the queries, we can use Material views which can be used as a CTE almost but on the fly, Endpoints to query them as a URL. WOW!

   3. The optimisation was based on the clickhouse features only, not sure if it would have been possible without it.

Learnt

1. SQL Recursive CTEs

   1. We can define a recursive CTE by referencing the CTE within it

   2. We have a single row (could be multiple as well) as the base case

   3. Then we define the recursive part, by referencing the cte as the table we are fetching the records from with the data queried to it as the parameter.

   4. Used it to solve Day 14

2. FTS in SQLite (Full text search)

   1. I learnt how to write a query for FTS and construct like a index for searching across tables.

   2. This is efficient from the string comparison as we don’t have to define how to look it up, we just define what we want. The algorithm and the query planner does it efficiently for us without storing it separately on disk.

3. Golang might just be better than python for writing LLM generated code

   1. Its simple has a type system. So it makes it easier for LLMs to generate valid code with correct checks in place.

   2. I need to experiment it with to understand the nuance this has.

Tech News

• Meta Buys Manus

• Google now allows you to change you gmail address

Leave a comment

Share Techstructive Weekly

Thanks for reading Techstructive Weekly! Subscribe for free to receive new posts and support my work.

Well! That was a thumping start to the year 2026. I have bright ideas and a new canvas to paint. Looking forward to have things running and working in my favor over this year. After a slog and slump for 2 years, its time for me for redemption. I can see a hope, hopefully you can too. If not, you will soon.

Happy New Year!

Happy Coding :)

It was winding down week, 2025, ending slowly, the biggest irony. The year was full of fast-paced, unprecedented models just dropping here and there. It was hard to keep up with the progress. It was getting overwhelming, no one is used to such a level of information. I was dreading with so much power in our hands. I couldn’t handle it and refrained it and thought it was better to focus on learning new things. I kept using AI tools at work, not by choice, but initially through force but then through necessity.

People made developers go fast, but they realised,

• Churning code was never the bottleneck

• Solving problems and understanding the business needs was the core goal.

Yet, people thought and here we are with tools like Cursor, Claude Code, and 10 other CLI tools. I learnt the hard way to leverage this tools, to use only when the code was throw away, some scripts, and getting insights from logs. But never on actual bugs and features, it was draining, lacked the joy I get from actually making it.

So, at the end of 2025, I am a decent developer who cannot use AI tools. Yes! give or take, I would love to learn more in 2026.

Quote of the week

Yes! That is what I believe in this year. I have stopped thinking it seems. If in 2026, we move very fast without thinking, I need to change for the better. The directions we take, everything is a decision, but on what ground, on what thought. If the thought are not 90% yours, you are not thinking enough.

Think, don’t let AI do it.

Created

• Advent of SQL 2025 in SQLite

  • Day 11: Behavior Score

  • Day 10: Misdelivered Presents

  • Day 9: Evergreen Market Orders

  • Day 8: Product Catalog

  • Day 7: Polar Express Mixin

  • Day 6: Days of Delight

  • Day 5: EchoTrack Wrapped

Read

1. You’re not burning out, you’re essentially starving

   • This was a good one

   When you truly chase your highest potential, everything you thought was burnout will melt away. Because you weren’t suffering from too much work, you were suffering from too little truly important work. Like a boy who thought he was full until dessert arrives, you’ll suddenly find your hunger return!

   Some really good points

   Pause once a month to make sure you’re still on the right track. Stop once a year to triple-check you’re on the right track. But never get off this path towards your highest potential. Anything else will starve you existentially

   This is true

   We’re optimizing for less suffering instead of more meaning.

   Yes

   I woke up today so excited to get to work thinking it was Monday morning already. Instead of jumping right into it, I spent all morning making breakfast and playing with my kids, then wrote this post. When I’m writing about something personal, 1,000+ words can easily flow for me in an afternoon.

   Just read the post!

2. Don’t become the machine

   1. This is well put.

   2. I kind of hate this argument.

   3. Why are we comparing ourselves to machines in the first place? We can grind, but with thinking what actually we are doing.

   4. Because I equate grind to consistency, it sometimes feels like grind, and we need to overcome that emotion of letting it overtake us. But most of the days, the grind is a joy, we do it because we feel like doing it.

3. How I left youtube

   1. Man that was a good read!

   2. I resonated with this a lot

      This duality is exhausting. It forces you to lie by omission to people you respect. You can’t tell your team, “I can’t take that ticket because I need to study dynamic programming.” You just have to work faster.

      I respect people above and behind me, but I too needed to move in life, support the things I was responsible for, get out of the grave situation I was pushed into. For that, I took some decisions, which I tried for, but nothing came off it, I wasn’t quite sure about the switch and left it when the offer came. Stranded here. I am feeling good here, but if I am not in another company by the end of 2026, something is wrong with me.

      Let’s see!

   3. Good lessons

      • Don’t say: “I tweaked the YouTube watch-time algorithm using X variable.

      ”

      • Do say: “I optimized a high-throughput distributed system to prioritize user retention metrics, reducing latency by 150ms through a custom caching layer.”

      Man!!

      My final conversation with my manager was heart-wrenching. I had prepared a script, anticipating a counter-offer or a guilt trip. Instead, I was met with soft and understanding empathy. 

      Yep

      The interview fatigue is real, and the conversations are hard, but the clarity you gain on your own value is worth the struggle.

   4. Got to go through it once and then there would be no stop for growth.

      Grass looks green on the other side always! Damm

4. Text editors should be worse

   1. Ok! I agree and disagree. You need to have a zen mode in your editor, which just is bare bones, and one for full fledged stuff like LSP, AI-auto-complete, syntax highlighting and what not.

   2. Editor is just a tool, it can’t code on its own(in 2025, still needs prompting), similarly to use it, it needs preferences.

5. Structured output can create fake confidence

   1. Spicy take and true! Somewhat true

   2. If your task is complex to get things out from image, or understand the context, it might hinder the quality.

   3. But if your task is to simply do something straightforward tool calls, structured output beats everything.

6. Writing HTML by hand is easier and cheaper then debugging your SSG in 2025

   1. Wow! We are moving at a pace where generating html from LLM is getting easier (not cheaper yet!) than generating it by code, whew! What a time to be in.

7. Guide to Local LLM Models

   1. Ok, the VRAM and RAM is somethign is quite critical. If you have less RAM and much VRAM, its no use, you need to have sufficient RAM in order to run a good enough model, VRAM wouldn’t handle it.

8. 13 Years of Rust and the birth of Rue

   1. I see this a lot! People creating something that they wanted but didn’t had the mental energy for.

   2. I see it as draining rather. I can’t watch it write code for me, its a dreading feeling to be in for larger durations.

9. Vibe Coding is broring

   1. It is pathetic, really. Watching it clog some code and done. Sigh what is left out then, to read code? Who loves it.

   2. Vibe coding is cool and good if you just want the product in your hands, but if you care about the craft then please write it.

Watched

1. Bublesort is useful

   1. Yes, this is kind of nuts

   2. Buble sort is the lowkey high value thing to learn and know of.

   3. VIsualizing any sorting algorithm really makes you understand the flow better and it clicks almost everytime.

2. The Fundamentals by Kelsey Hightower

   1. Ahh! How many people will say this, but yet we can’t follow it

   2. Everything boils down to the fundamentals, having the basic thing to understand when something goes wrong. Rather we make it complex in order to be percieved as smart and even oversmart.

3. Will Turso be the better SQLite? Interview with Glauber Costa

   1. This was a great interview. I love the mentality.

   2. If software built with community in the Linux community can sustain after almost 3 decades, then why can’t a embedded database like SQLite can?

   3. Turso is Linux Community but for SQLite (minus the toxic leadership)

   4. Pekka is a great, humble and smart leader to be leading the Turso, SQLite rewrite in Rust.

   5. I want to contribute to SQLite, but it feels I don’t know enough everytime I touch it, also I started learning SQL for this. I have gone so far and now there is no way I am turning back.

   6. I had one itch for geospatial exploration in SQLite for Mumbai city. This weekend might be the time to do it, maybe next year.

Learnt

• Using Tool calling in Google Gemini API

  • We can pass the tool as Code Execution block and it can essentially work as an agent in the api. This is a superpower to have.

  • Can imagine people creating workflows and all sorts of things with the api in gemini, and gemini, kid you not is a really good model, like it can just do things. (Not complex things, but simple things, it can do really well)

• I played with Grok Imagine

  • Oh boy! These image and video models are getting really out of hands. I just uploaded my photo and boy came a introduction about me, like a one sentence greeting, but it was scary that it can do that, that quickly.

  • I underistimated how quickly these models will evolve, we might plateau out eventually, but still the progress made is mind boggling.

Tech News

Well its Christmas and end of year.

God! Dam! This AI labs have learned something from last year. We don’t have groundbreaking models now! 2025 was a rollercoaster.

For more news, follow the Hackernewsletter (#775th edition), and for software development/coding articles, join daily.dev.

But here’s my wrap in 2025

• Wrote 52 articles from #23 all the way to #74

• Wrote 50+ articles on SQL on #sqlog

• Learnt about SQLite and solved 15 advent of sql #advent-of-sql-2025

All in all around 120 posts on my blog #blog

See you next year!

Merry Christmas and a Happy New Year!

(oh before that, there will be a 2025-yearly-review post)

Happy Coding!

Leave a comment

Share Techstructive Weekly

Thanks for reading Techstructive Weekly! Subscribe for free to receive new posts and support my work.