How do I stop AI from sending emails on its own?
Set each agent to draft or review, send every email through one approval queue, and keep a single switch that stops them all.
Want to see it first? Book a demo
The short answer
- You stop AI from sending emails on its own by setting each agent to draft or review mode, so every message waits in an approval queue until a person approves it.
- Guardrails sit under the queue: days between touches, daily caps per mailbox and domain, a business-hours window, and a suppression check at the moment of send.
- One kill switch that stops every agent at once beats a separate toggle per agent that someone has to remember to flip.
- SalesCrew routes every external send an agent proposes through one approval queue by default, and its kill switch on the Agents page stops every agent at once.
A toggle answers the wrong question
An autopilot switch per bot tells you whether an agent is running. It doesn't tell you whether a person read the email before it left, and that's the question your prospects care about.
The gap shows up fast once you run more than one agent. One is off for replies but still sending follow-ups through the same mailbox. You find out from a prospect. A review queue fixes the order of events instead: the agent drafts, a person approves, then it sends.
Set up human approval in five steps
The same order works in any tool that has agent modes and a queue. The labels below are SalesCrew's.
- 1
Choose a mode per agent
Off does nothing. Draft writes and stops. Review queues the message for a person. Auto sends and logs, and drops back to review when confidence is low.
- 2
Keep every send on review
Email, LinkedIn messages and marketplace proposals default to review. Leave them there until the drafts are consistently right.
- 3
Let internal work run
Research, tagging, scoring, summaries and logging run on auto. Nothing a prospect sees happens without a person.
- 4
Set the guardrails
Pick the days between touches, the daily caps per mailbox and per domain, and the business-hours window.
- 5
Know where the switch is
The kill switch on the Agents page stops every agent at once. Its MCP twin is agent_set_mode.
Six controls that stop an unwanted send
| Control | What it prevents | How SalesCrew applies it |
|---|---|---|
| Review mode on external sends | An email leaving before a person reads it | Default for every agent's email, LinkedIn message and marketplace proposal |
| Days between touches | The same contact hit twice in a short window | At least 3 days between touches per contact, across every cadence |
| Daily caps | A burst of sends that hurts deliverability | Per mailbox and per domain, plus a per-cadence cap |
| Business-hours window | Messages landing at 3am | Sends only inside business hours |
| Suppression at send | An email to someone who unsubscribed after the list was built | Checked again at the moment of send |
| One kill switch | Any agent still acting after something goes wrong | One control on the Agents page stops every agent |
Defaults as SalesCrew ships them, September 2026. Each one is editable in Agents, then Guardrails.
What approval looks like in SalesCrew
SalesCrew puts every agent's proposed send in one approval queue, grouped by agent. Each item shows the draft, the evidence behind it (the thread, the lead score, the knowledge base sources), the confidence and the cost. You approve it, edit then approve, or reject it with a reason. That reason feeds the agent's next run.
You can also skip agents and write with help. Draft with AI writes a reply inside the inbox composer, and you send it yourself. The inbox agent goes further: it sweeps new threads every 15 minutes and drafts replies for the queue. It starts switched off, and you pick draft, review or auto. Every tool call from a person, token or agent lands in the audit log. More on the agents and approval queue.
Check every path to the mailbox, the agent included
Questions
- How do I set up human approval for AI-written sales emails?
- Set the agent that writes the email to review mode, so each draft lands in an approval queue instead of the outbox. A person then approves, edits or rejects it. In SalesCrew every external send defaults to review, and you set each agent's mode on the Agents page.
- Is a per-agent on/off toggle enough?
- A per-agent toggle stops one agent from running, but a running agent still sends without review. With 3 agents you'd need to remember 3 switches when something goes wrong. One review queue for every send, plus one kill switch for every agent, covers both gaps.
- Does review mode slow the team down?
- Review mode adds one read before a message leaves. Tagging, scoring, summaries and logging can still run on auto, so the review lands only on what a prospect would see. In SalesCrew, create-deal and book-meeting actions can move from review to auto after a clean week.
- Can the AI still draft replies if it can't send them?
- Yes. Draft mode writes the reply and does nothing else, and review mode queues it for a person. SalesCrew's Draft with AI writes a reply inside the inbox composer, and the inbox agent sweeps new threads every 15 minutes and drafts replies for review.