I wanted to see how far I can get running a strong coding model - m2.5 from minimax - in an agentic loop on my dgx spark. What's not in the post yet: it's running, I gave it a project, completely hands off, used about 1.2m tokens.
I have a lot more learnings from this than I can put in 1 post. So this is the framing. I'll drill deeper into topics I've explored later.