Lenny RachitskyHost of Lenny's Podcast | Author of Lenny's Newsletter

Why is AI so good at coding?

6 Apr 2026 · 1 min

Watch on YouTube

1:00

AI excels at coding because self-testing agents like Claude Code can run their own code, find bugs, and fix them instantly. This feedback loop doesn't exist yet in law or most other fields, where you can't automatically verify if something works.

Chapters

The real unlock was these coding agents. It used to be you'd ask ChatGPT for some code and it would spit out some code, and then you look at it and you have to run it and test it. The tools like Claude Code and OpenAI Codex, they take that step for you. They can generate the code and then they can run the code themselves and see if it works. And if it doesn't work, they can fix it. You can watch them fix their own bugs. And that made all of the difference. We went from they spit things out that look good, but who knows if they work, to they spit things out and then they test the things. And see if they work or not. And that loop, I think, is still unique to coding. I don't think you're gonna be able to hook up an agent that goes and does a lawsuit and argues in front of a jury. It's a lot harder to get that sort of self-correcting loop in place for other fields. There's a lot of effort going into things like finance, right? Finance is a field where you can at least do a lot of this evaluation directly. And the open question for me is how many other knowledge work fields are actually prone to these agent loops?