展示 HN:治理的 AI 投资组合——用于生产中自主系统的准入控制
大多数代理失效并不是模型失效,而是组织记忆失效。<p>我为在事件后接受审计的团队构建了一个开源控制平面架构:
- 决策是合同(而不是日志)
- 治理是准入控制(而不是官僚主义)
- 证据能够在团队更替中保存<p>包括:决策合同、持续集成门、审计级文档和三个受管控的变更胶囊。<p>GitHub: <a href="https://github.com/lexseasson/governed-ai-portfolio" rel="nofollow">https://github.com/lexseasson/governed-ai-portfolio</a>
查看原文
Most agentic failures aren’t model failures — they’re organizational memory failures.<p>I built an open-source control-plane architecture for teams that get audited after incidents:
- Decisions are contracts (not logs)
- Governance is admission control (not bureaucracy)
- Evidence survives team turnover<p>Includes: decision contracts, CI gates, audit-grade artifacts, and 3 governed change capsules.<p>GitHub: <a href="https://github.com/lexseasson/governed-ai-portfolio" rel="nofollow">https://github.com/lexseasson/governed-ai-portfolio</a>