Comment by sigmoid10
1 month ago
IMO with the latest generation (gpt codex 5.3 and claude 4.6) most devs could probably be replaced by AI. They can do stuff that I've seen senior devs fail at. When I have a question about a co-workers project, I no longer ask them and instead immediately let copilot have a look at the repo and it will be faster and more accurate at identifying the root cause of issues than humans who actually worked on the project. I've yet to find a scenario where they fail. I'm sure there are still edge cases, but I'm starting to doubt humans will matter in them for long. At this point we really just need better harnesses for these models, but in terms of capabilities they may as well take over now.
> most devs could probably be replaced by AI. They can do stuff that I've seen senior devs fail at.
When I read these takes I wonder what kind of companies some of you have been working for. I say this as someone who has been using Opus 4.6 and GPT-Codex-5.3 daily.
I think the “senior developer” title inflation created a bubble of developers who coasted on playing the ticket productive game where even small tasks could be turned into points and sprints and charts and graphs such that busy work looked like a lot of work was being done.
they are good at some weird problems - but also write some really bad code and sometimes come up with wrong answers.
That's why you write tests.
There are whole classes of problems that tests can't catch.