Thread

Yeah, I’ve noticed that too. It seems to struggle with layout logic and selector specificity, maybe it wasn’t trained deeply enough on front-end design patterns.

coming from claude sonnet and opus, grok fast is capable but kind of loses the plot after a few turns. i'm not sure if this is a fundamental model size limitation, or just an artifact of how it was trained

Yeah, I tend to use Claude to plan, and then grok fast to implement. If I really get stuck then throw it to GPT. I have found however that I can often just run in code mode with grok code fast. But just last night and this morning I was working on some front end stuff that seemed fairly easy to me, that completely stumped grok fast. Had not noticed this before.

what kind of tool? an ide like cursor, or a tool using agent like goose / claude code?

roo code