I keep seeing people say we should treat AI agents as junior developers, but I can't do that.

Because I treat juniors as future seniors that I get to help build. But the current AIs cannot ever become seniors, and if a future gen could, then the rest of us are even more fucked.

So no, that's not a good mental model.

#AI #LLM #phptek #phptek26

@Crell and seniors also learn from working with juniors… sometimes juniors
know stuff that seniors dont, sometimes juniors not understanding stuff is not a sign that juniors need to be trained, but that the stuff they dont understand needs to be made easier to understand. all this will be lost if we follow/ join the #AI-#LLM sect.
@jpaulgibson @Crell Ehh, depending, if an LLM can't make sense of a codebase, that may be a code smell indicating that cleaning up the code is worthwhile. And I've learned (after verifying of courese) things about how frontend stuff works because a bot threw some bit of knowledge at me. Some of the better models hallucinate less.

@jpaulgibson @Crell There's also a measurable improvement in performance when code is documented. Which is good idea to do anyway versus keeping things in folks' heads!

The things that make code better for humans generally makes stuff better for LLMs, and the latter are provably (error bars apply, see nondeterminism for details).

The idea that LLMs will wade through the slop without slowing down is incorrect and wasteful. And you can use the tools to clean up the slop. Just takes more effort.

@Crell its a mistake to even treat today's ai as ai. because actual ai would be sentient enough to make you feel bad for treating it poorly as it stares you in the face with unblinking eyes.

this is just a natural language front end for a vector database and nothing more. sql but worse since rng is a primary feature.