A place for developers to talk about building on Farcaster.

/dev

haardik.dev | unreasonably optimistic, relentlessly curious

The more I push it the more I’m disappointed (as is the case with every LLM). Just stupid lazy mistakes. Horrible quality in even the best cases. No “agent workflow” comes even close to the accuracy of tab completion which isn’t unique to cursor

And its slow as hell

I think it depends on how youre using it. It generally doesn't work well on big projects or when youre using terminology it may not know. I use it mostly for bootstrapping or early stage projects, or tell it to explain a code architecture for smth I wanna do without editing any of my files - works well for those use cases

But big projects are the real ones. Anyone can take a project from 0 to 1, maintaining and improving legacy systems is the hard part

Oh for sure - im just telling you what its actually good for vs what its not right now

I think the assumption is LLMs will just get better over time