Got roasted by a senior dev for trusting AI code suggestions blindly

I was using GitHub Copilot to generate a bunch of API endpoints for a side project and a senior dev at my meetup pointed out I had no idea what half the code actually did. He literally said "you're just copy-pasting bugs you don't understand yet." Now I manually review every line Copilot suggests, but it kills my speed - has anyone found a good middle ground for trusting AI assists?

3 comments

3 Comments

foster.tessa1mo agoMost Upvoted

I used to be all about trusting Copilot blindly too, but that senior dev's point really hit me hard. I had a situation where Copilot suggested a rate limiter that looked perfect but was actually skipping validation on certain edge cases, and I nearly shipped it. Now I treat AI suggestions like a rough draft that I have to fully rewrite line by line. My middle ground is using AI to get the structure started but then rewriting every function myself, which actually helped me learn way more about how the code works.

gavin4691mo ago

That approach just slows you down without fixing the real problem.

jana8811mo ago

Lmao that rate limiter story is gold. I mean honestly the idea of blindly trusting a machine to write production code is wild when you think about it. My favorite was when Copilot suggested a authentication middleware that checked for a token but never actually verified it was valid. Just said "if token exists, allow access." Like thanks buddy, that's basically an open door policy. The senior dev was right though, it's not really about speed or trust. It's about knowing that if you ship broken code, the AI isn't going to be the one getting the angry 2am slack messages.