生产级 AI Agent 的 Harness：控制面、不变量与审批边界
原标题：Presentation: The Agent Harness: Control Planes, Invariants, and Approval Boundaries for Production AI Agents
AI 摘要
OpenAI 核心数据与 AI 基础设施团队成员 Vinoth Govindarajan 在演讲中提出「Agent Harness」概念，主张生产级 AI Agent 应由模型之外的系统层负责控制、状态与审批边界。他以 OpenClaw 为公开案例，指出「静默成功」比崩溃更危险——用户看到回复成功，但持久化记录出现空洞，导致后续上下文基于不完整记录推理。他给出三条原则：状态需有唯一归属与重放路径、变更需有序提交、需用回执证明实际执行而非仅凭对话记录。
正文节选
Transcript Vinoth Govindarajan: This talk became possible by reading a lot of GitHub Issues over the weekend. I know based on your personality, either call it research, therapy, or cry for help. I know when you build production systems, you know the truth. Most of the important lessons lie in these bug reports and issues. Let's start with one sentence. A user saw the reply. The system forgot it happened. This is a failure shape I want to start with. Not a model hallucination, not a bad answer, n