Why my AI agents can't spend, send, submit or publish
说明与用途
在一家小型线索生成机构中,AI 代理可以进行研究和草稿编写,但在未获得书面批准前,不能自行消费资金、发送消息、提交表单或发布网站。四个“门”规则和审批队列的系统可以防止错误,但会减慢工作流程。
问题
AI agents in a small lead‑gen business must be prevented from making irreversible or costly actions (spending money, sending messages, submitting forms, publishing content) without human approval.
结果
After one week, 154 businesses were audited, 78 mini‑audits produced, 10 emails sent, 2 rank‑and‑rent sites and 2 referral sites live, but no revenue was generated. The gate system prevented accidental spending or publishing.
工作流
- 1. Define four gate rules (Spend, Send, Submit, Publish) and a source‑link rule.
- 2. Prepend these rules to every agent’s instruction set.
- 3. Agents perform tasks; when a gate is triggered they record the step under "decisions needed" and continue other work.
- 4. A Chief of Staff agent aggregates all blocked steps into a single queue.
- 5. The human reviews the queue at fixed times, approving, modifying, or rejecting each item.
- 6. Approved actions are executed (e.g., sending emails, purchasing, publishing sites).
证据
Every agent gets the same short block of rules pasted at the top of its instructions, above everything else:
When an agent hits a gate, it stops that one step, writes it down under "decisions needed" in the daily debrief, and moves on to other work.
After the first week the queue holds a fair amount: 18 drafted emails for a second outreach batch, 50 drafted pitches to other agencies, and 5 full audits built ahead of time.
The split after a week looks like this. The agents: market research, competitor notes, domain checks, site briefs, page generation, sources files, prospect lists, homepage audits, one-page reports, email drafts, the verification re-check, price research. Me: picking between shortlisted markets, approving each purchase, reading and sending each email batch, deploying each site, and anything that needs my legal name, tax details or bank details.
- 记录日期: 2026-10-11T09:56:09.117Z