The question is, if we trained an LLM on human's piloting Claude Code.
Then had that LLM pilot Claude Code
Is that enough to build a recursive self-improvement loop?