How-toMulti-source
Anthropic 讲解 Agent 权限设计：提议与执行分离
Why we picked it
Anthropic 的人用 83 秒讲明白一个实用思路：让 Agent 只出方案不动手，执行前加人工审批，改 Agent 前先跑 evals。
Anthropic 的 Keon Shahab 在 AI Engineer New York 演讲中用一个 83 秒的例子说明：Agent 可以提出交易修改方案，但没有权限直接执行。执行前需要经过人工审批环节，并通过 evals 测试验证。这种提议与执行权限分离的做法，是控制 Agent 风险的一种具体方案。
Original · AI Engineer
An agent can propose a trade amendment without being allowed to apply it. Keon Shahab of Anthropic walks through permissions, human approval and evals. 83 seconds from AI Engineer New York. What do you test before changing an agent? #AIEngineerNY Your browser does not support the video tag. 🔗 View on Twitter 💬 1 🔄 0 ❤️ 2 👀 586 📊 1 ⚡
