Human control of agentic systems

OpenClaw’s fully agentic forums are a fun demo on the surface, but they’re also an early peek at a pretty weird failure mode for vibe coding. Once you let agents recursively spin up infra, you don’t just get messy codebases. You get an entire socio-technical system that no one can really audit. The usual governance primitives we rely on as founders start to blur: - Ownership: who’s actually responsible when an agent-of-an-agent deploys a thing that another agent configured? - Review: what does code review even mean when most of the changes are machine-to-machine and happening faster than your team can read a pull request? - Rollback: revert to what, exactly, if the system is continuously mutating itself based on feedback loops you don’t fully understand? Can humans still exert legible, enforceable control over what’s running in production?” And can we do it in a way that passes a security audit, a board meeting, and maybe one day a regulator?