[koiueo]

> We’ll miss the sleepless wrangling of some odd bug that eventually relents to the debugger at 2 AM.

I'll miss it not because the activity becomes obsolete, but because it's much more interesting than sitting till 2am trying to convince LLM to find and fix the bug for me.

We'll still be sitting till 2am.

> They can write code better than you or I can, and if you don’t believe me, wait six months.

I've been hearing this for the last two years. And yet, LLMs, given abstract description of the problem, still write worse code than I do.

Or did you mean type code? Because in that case, yes, I'd agree. They type better.

[cmiles74]

I am not confident that AI tooling can diagnose or fix this kind of bug. I’ve pointed Claude Opus at bugs that puzzle me (with only one code base involved) and, so far, it has only introduced more bugs in other places.

[koiueo]

I'm not saying it can btw. I'm arguing for the opposite.

And for the record, I'm impressed at issues it can diagnose. Being able to query multiple data sources in parallel and detect anomalies, it sometimes can find the root cause for an incident in a distributed system in a matter of minutes. I have many examples when LLMs found bugs in existing code when tasked to write unit tests (usually around edge cases).

But complex issues that stem from ambiguous domain are often out of reach. By the time I'm able to convey to an LLM all the intricacies of the domain using plain English, I'm usually able to find the issue myself.

And that's my point: I'd be more eager to run the code under debugger till 2am, than to push an LLM to debug for me (can easily take till 2am, but I'd be less confident I can succeed at all)