Overheard a dev say LLMs are just autocomplete, my take after 3 months

I was at a coffee shop in Austin last Tuesday and heard two engineers joking that LLMs are just fancy autocomplete. I get the skepticism, but I've spent 3 months using one to debug legacy Python code at work, and it caught a race condition I'd stared at for 6 hours. Sure, it's not thinking, but neither is a calculator, and we still call that a tool. How many of you actually use these for real debugging, or is it all just draft emails and blog posts?

1 comments

Log in to join the discussion

Log In

1 Comment

holly_garcia1mo ago

Have you tried feeding it the stack trace and just letting it talk through the logic out loud? Because that's been my saving grace more than once. I think people who call it autocomplete are underselling how useful it is to have something that never gets tired of asking 'what if you tried this instead'. It's not magic, it's just a second set of eyes that doesn't have ego or a timeline. Your race condition story doesn't surprise me at all, I've had it point me to a variable shadowing issue that I'd have bet money wasn't the problem. It's a tool, sure, but so is a debugger, and nobody sneers at those.

2