That last paragraph echoes my recent experience of dealing with LLMs at work (we've finally been told to use it despite my objections).

The code works, I guess, for whatever immediate things the person told the LLM, but there's A LOT of code and it's rife with strange architectural decisions. It's taking me forever to get through reviews and I've become a bottleneck. I see something strange in the code and now I have to go through and try to reverse engineer the decision myself, because the person who opened the PR lost sight of the details.

Maybe I'm worrying too much about the details now since the next iteration of LLM output will just pave over it. Maybe code quality doesn't matter on an LLM driven team, but it sure does make review difficult.

This whole thing feels like we are sacrificing the future for perceived short term gains. Or maybe those gains or real, I really don't know anymore.

I can't shake the feeling that this whole thing is like building a bridge without actually verifying the members. It might seem ok now, but is the bridge going to collapse in a few years because of an improper joint?

If “Adam Thornhill” wants to claim 40 years in the industry, they should use an author photo that isn’t 40 years old.