The approval gate: a pattern for AI agents that write to production systems

说明与用途

审批门模式将决策与执行分离，让人工批准 AI 代理提出的更改，从而降低关键生产系统中的错误风险。

问题

AI agents that write directly to production systems risk corrupting critical data, eroding trust, and incurring high cleanup costs when writes are incorrect.

结果

The approval gate pattern reduced the impact of erroneous writes, improved auditability, and restored trust among revenue team users, enabling continued automation.

工作流

• 1. Agent generates a proposed action with diff, evidence, and idempotency key.
• 2. System validates the proposal against target schema and permissions (preflight).
• 3. Proposal is queued for human review.
• 4. Reviewer approves or rejects the diff within seconds.
• 5. Executor applies the approved change and logs the full envelope, including inverse action for reversibility.
• 6. System records audit trail for future reference.

证据

At 95 percent action accuracy, one in twenty writes is wrong, and those writes compound into a database humans use to make decisions.

来源

Show the reviewer from and to. A human can approve or reject a diff in about two seconds.

来源