Techstructive Weekly #103
Reading books instead of tech things, understanding and solving problems with AI, among the other things watching and learning from the week 12th to 18th July 2026
Week #103
Another week with love for solving problems as a developers, just the leverage has changed. Instead of hand coding, we hand it off to an agent and focus on the broader problem. That might be not sustainable as I have said before. But if the problem is big, then it becomes just like the coding challenge, you don’t want to get bugged down by small pesky bugs on code rather think on the bigger problem.
Continuing the momentum into the next week.
Quote of the Week
“The ocean comes to the shore every minute, yet it gets pulled back. But it never gives up nonetheless, it unknowingly creates the rhythm of the ocean”
— Not quoted by anyone, just saw it somewhere similar to this quote
Its so deep yet simple. Ocean keeps washing its water over the shore, and has to recede it back, like someone trying to get something yet getting rejected or failing. But it doesn’t stop it, it keeps on trying day in and day out. Probably til the earth is there, and there is water in the oceans. This is what we refer to the rhythm of the ocean right? So beautiful that for many people including me, its a therapy to see and listen the ocean at the shore.
Life is like that, out of small grit, you develop a habit, and that habit becomes your character and the personality with which you develop a identity and stay in memories of people. Just keep doing what you love, consistently, no pressure, some day you’ll too be called a rhythm of ocean.
Read
Pydantic Blog: The human in the loop is tired
Take my money for writing this post!
“everything I write goes into some AI black hole. There's no person on the other side actually learning anything”
So real. The part of interacting with an agent is just in-humane. It doesn’t give any skill boost. It makes us feel detached from the world.
It feels good that an organisation which is probably at the forefront on the AI revolution, Pydantic is writing such human heart touching posts. Makes them valuable then ever.
Doug Turnbull Blog: Why write code in 2026
In the era where people defend with “I read the code”, this blog post slows down and answers a deeper question. Why write when nobody is even reading the code.
Writing code has and will continue to give me clarity along the way. But AI has made all of that clarity to be required upfront which is a bit taxing on the brain.
Like imagine starting to work on.a feature, you start with the boilerplate code and then start to write something, then you realize oh, that is not right, we need this too, we won’t require that I think. And you get the idea, the clarity comes on the way and not staring at the agent after leaving it with a one line description of the task.
Hugo Daniel Blog: The end of creativity
So true. Everyone will now want to use AI to be lazy.
Creativity is something that you don’t get from letting someone else do it, its making mistakes and breaking things. But AI has made it even harder to explore creativity in some sense. It has expanded the imagination and what we can do with the kindle of a thought but beyond that, there is no bridging of that gap of fear to action.
George Hotz Blog: I love LLMs, I hate the hype
This is cool. Some big people are stepping up and raising the alarm on the hype noise.
This is a great thought in here “You have to be really careful, they can increase cognitive fatigue, and all the vibe coded stuff is still slop”
Perfectly put. Developers still need to hold the leash for what gets shipped.
Watched
GPT Models in Claude code are better than in codex?
This is wild. The reason was hidden the codex system prompt is too good. I mean bad. It sounds cheesy, hard coded times, frontend design is a mess. Ouch! that hurted.
I had used codex extensively in the month of March to May, but then it was all limited and monthly quota got exhausted in a single prompt or I think they changed some things on free tier, not sure. But haven’t used it much since then.
I regularly use Claude code and Cursor with opencode in the mix if I need something quick and dirty scripts to get out of my way.
GKCS: 20 Concepts to know in AI Engineering
No bullshit guide on the concepts. Straight to the point and precise. It is a great video to kindle the spark that there are a lot of things to explore and do even though AI can write code. Writing code is just one part of it.
I liked the explanation of top p and top k and temperature. That separation of idealogy makes it so much valuable.
The explanation of reinforcement learning, self supervised learning were all great to understand and know at least.
GKCS: Current State of AI Engineering
This is really eye-opening. Engineers are not going anywhere. But surely we need to adopt. Design principles still remain vital.
Interviewing is a skill that is now taking out from the code to the high level design and the details too. This is perfect. I don’t like LeetCode solving but like talking about specific design decisions I made while making projects or working in a job. That feels human.
Grant Sanderson on Dwarkesh Patel Podcast
Just math curious. Like the who rather than what was a interesting one.
I like his approach on learning and teaching. It also gives me hope, that humans will do math just because they want to. Its such a silly thought but a lot is said in that.
Standup Math by Matt Parker: Has AI discoverd a new maths, vibe mathing?
OK, AI can write math proofs now. So what, we will have more math to do! Amazing right? right? Maybe yes, let’s see.
Average Primeagen video be like reaction by Primeagen
This was really funny. Straight and vertical lines are no joke to do, and on live stream, its a next level.
Just a great video to watch amidst all doom and gloom of AI.
Learnt
Writing working software is no more the goal or never was
I was working on a feature that adds a agentic block on a workflow.
Yeah! That sounds insane right? Adding llms as a brain as a block in a workflow. Kid you not, its the only solution when you can’t find relevance with a single llm call. You need to bring the pieces together and let the llm think and act.
For this to work, you don’t need working software, you need a clear generic goal. And defining that is no trivial. I am still in the hunt for that.
Tech News
Kimi K3 is the best Frontend model: Not just open weight, but out pacing leading frontier closed source models too. Moonshot RELEASE BLOG
For more news, follow the Hackernewsletter (#801st edition), and for software development/coding articles, join daily.dev.
That’s it from the 103rd Edition of techstructive weekly. I hope you found it helpful, and relaxing. If not please drop any suggestions, feedback or discussion about certain things you want to in the comments or drop me a message on my socials.
Thank you for reading,
Until next week.
Happy Coding :)