My experience with AI

I'm not a vibe coder. I have been writing and reading software code for decades and that left a lasting impression on me. I won't commit code that I personally don't understand, even if it means that it slows development. However, I think that modern LLMs are technological wonders with huge potential and I use them in my work. Fortunately, working with agentic systems is not unlike leading a team of junior developers and I have some experience with that. Make and maintain a plan, split work into manageable chunks, cover everything with tests, make key decisions and intervene if necessary, create good documentation to make it easy to trace your design. Any expert in software development knows the value and importance of all that, regardless of whether the implementation of all that is made by LLMs or humans. I'm still learning the best ways to work with AI, though.

I also found that LLMs are invaluable tools for search. No matter how smart and knowledgeable you are, you can't be an expert in everything. Doing research, reading books and asking questions have always been an essential part of the job for a good software developer. Modern AI has made this process so much faster & better. This observation applies to searches within large code bases as well.

Finally, LLMs are pretty good at software reviews, especially for small issues that comprise the majority of comments in most pull requests. It is a pain to integrate that with old corporate systems, though.

This web site in particular was developed with Copilot and (mostly) Grok 4.6.