How continuous learning shapes effective software development with LLMs
Link: https://martinfowler.com/articles/llm-learning-loop.html
This is a true analogy. This is highlighting about the world relating software development with a assembly line.
Source: techstructive-weekly-67
Link: https://brandonharris.io/Garbage-Pail-Code/
I would have loved this post, but the AI-generated image just threw me off a bad impression. But it was chess, how can I resist this. I read this, and felt good. A human admitting his mistake. Going all in AI and failing hard, then going again but from the lessons learned from the fall.
Source: techstructive-weekly-67
For as long as I have published my books, one of my overarching goals was to give credit to those who actually invented the hardware and software that we use. I have spent 10,000+ hours to create an accurate record of their work but I'm not …
Link: https://inventingthefuture.ghost.io/mr-tiff/
It was just a pleasant read. No controversy, no fluff. Just two humble and honest humans doing their job. This world would be so beautiful if each human did its job correctly, just correctly, not exceeding the expectation or performing a all-nighter. Just completing his or her job correctly is so under-rated.
Source: techstructive-weekly-67
They're like riding a bike: easy, and you don't get it until you try.
Link: https://fly.io/blog/everyone-write-an-agent/
A good one. You need to experience the thing in order to have an opinion. That’s quite obvious. But people coming on conclusions about AI and LLMs don’t quite think that way. They do one thing and generalize their opinion. Generalization of experiences is not good. I have used Cursor, cursor-agent cli, gemini cli, and amp cli free version. I can have opinions about them. But I cannot about Claude Code, Codex, and the bazillions of those CLI agentic models.
Link: https://bytesauna.com/post/coding-vs-software-engineering
They just don’t write code, they vomit and throw code at you. To review. It can’t develop intuition for what to debug, what to test for. It will do whatever it takes to make it work, it could even mean to delete the test and change the code to pass the test. Keep it away from your zone.
Source: techstructive-weekly-66
Link: https://youtu.be/rgiuaJbyUyU
: Prime Reacts Agreed. I have never used the .md files for AI Agents, it doesn’t listen to the main prompt. How can it listen and adhere to the other file prompts? People just make it look sophisticated, but it really doesn’t need it. The Tab completion and one line or word completion is good, even a great time saver, but generating large chunks of diffs, nope. Thank you. I still do it. The war against developers using AI and a developer hand chisling code, is true, you can’t compete.
Web presense of Marvin Hagemeister who is a core developer of preactjs
Link: https://marvinh.dev/blog/building-from-first-principles/
Agreed. Jargon creates back and forth between people. It leads to more confusion and bugs. Make it simple, you don’t need a 10-document-long API spec, a simple line can serve the purpose if the intention and the why is clear. Keep it simple and studpid, this is a well known advice and it should be injested in AI workflows, they create bloat and make it impossible to review the complex tree of code.
Link: https://www.geoffreylitt.com/2025/10/24/code-like-a-surgeon
Knowing what to hand off and what to roll your sleeves . Knowing that is cirtical. I think it comes off from experience and time, you will fail in making decisions, but it would make you aware what you are capable of and what will take time.
Source: techstructive-weekly-66
Link: https://youtu.be/VgtfpTGJmHQ
This is fast, blazingly fast. The model I don’t know but the speed is truly fast. If the intelligence is not degraded from gpt5, I think its a good option to have.
Source: techstructive-weekly-66
Link: https://youtu.be/XhJYNSssqrA
Brain rot. This is hillarious how reasoning, long context and short form content can alter.
Source: techstructive-weekly-66