I've been watching one of the devs on my team become a slightly worse programmer over the last few months, little by little, as he's been relying more and more on AI coding agents.
It's a real shame. He's actually a really good programmer. But as he's been using LLMs more and more, I see his code quality slowly declining. There are more bugs or just sloppy code in his PRs as time goes on. And his responses to comments on his MRs sound like they came straight from Claude Code.
I use some of the same tools he's using every day, but I'm using it to automate repetitive tasks and write my commit messages, and carefully reviewing everything written by the LLM. He, on the other hand, is writing out long multi-page prompts, handing them off to Claude Code, and then submitting the PR without careful review.
He thinks he's being more productive because he gets his PR out faster, but then spends a lot more time in review.
