Meta-commentary: these different AI agent systems (frameworks?) sure are good at racking up GitHub stars but don't appear to have any use beyond a fun demo. My impression is that, like AutoGPT, this will simply fall over with anything even mildly complicated. Maybe it's the direction we're ultimately headed, but I'm just far too skeptical given what I've seen so far, and I don't believe we just need the right orchestration of models to pull off useful general purpose AI agents.