Your Agent Didn't Fail. Your Harness Did. — Vinoth Govindarajan, OpenAI
The video discusses failures that occur not in the AI model itself but in the system around it—called "harness failures".
The problem is that agents can respond confidently based on outdated data while two processes overwrite each other without crashing, or when a tool is called but never receives a response. The solution is to view it as: the model proposes, the system executes, and a receipt proves it actually happened. The speaker shows common mistakes such as incorrect permissions, missing deadlines, and state that is never saved, and presents five questions to ask for every incident: what started it, what state did it inherit, what permissions were used, what ran, and what evidence remains.
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
More from AI Engineer
Inside Arca's Long-Horizon Agents That Run the Business - Aman Singh, Arca
AI Engineer 23 Oct
Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg
AI Engineer 23 Oct
Outside of tech there has been no financial payoff to US companies for using AI, now viewed as a margin-hurting cost center, as VC-backed firms burn more tokens
Techmeme 4 h ago
Source: Replit recently moved most AI spend from Anthropic to OpenAI, which gave 30%+ off in a deal worth $100M+/year, after Anthropic discontinued its discount
Techmeme 6 h ago