Theoretically, we can now task tools like Claude Code or OpenClaw to monitor a git repo, analyze the abstractions in completed work, and then autonomously generate new agents or skills capable of handling similar tasks in the future.

Is anyone successfully running a loop like this? I’m curious if anyone here has shifted the majority of their time from writing code to crafting these systems—essentially bootstrapping agents that learn from the repo history to build better agents. I'd love to hear from those pushing the boundaries on this.