Know
Presentation: The Agent Harness: Control Planes, Invariants, and Approval Boundaries for Production AI Agents
OpenAI’s Vinoth Govindarajan discusses why production AI agents fail beyond model hallucination. Using real-world case studies like OpenClaw, he explains the key principles of reliable agent harnesses: establishing explicit state ownership, serializing concurrent state mutations, scoping execution authority, and validating actions at the user-visible edge. By Vinoth Govindarajan