Disclaimer: This post was written May 2025, and the arguments apply to AI code capabilities at this time. The arguments around lack of competence are certainly likely to become less prevalent-while the parts about the desecration of the joys of programming, and fundamental human understanding of programming-are likely to become
The differences here are that ORM and web frameworks weren’t actively making the job harder and the sheer surface area of the problem.
If you fuck up with a framework or an ORM, it generally just fails to work, the magic internals might not be super helpful with their error messages, but such is the nature of the tradeoffs.
If you fuck up with an LLM you get something that generally compiles and looks like it should work, that’s much more of a problem for both you and anyone who then needs to go trawling through, looking for the issues.