Does SalesCrew send emails without approval?
No. An AI-drafted send waits in one queue until a person approves it, and one kill switch stops every agent.

The short answer
- SalesCrew does not let an AI agent send an external email without approval by default: every external send waits in one approval queue for a person.
- SalesCrew agents have four modes (off, draft, review, auto), and the default policy puts every external send in review.
- In the SalesCrew approval queue you approve, edit then approve, or reject with a reason that shapes the agent's next run.
- One SalesCrew kill switch on the Agents page stops every agent at once, and its MCP twin is agent_set_mode.
What waits for you, and what runs on its own
SalesCrew splits agent work by risk. Low-risk work runs straight away: research, tagging, scoring, summaries and internal tasks. So do the safety actions, like pausing a cadence when someone replies or suppressing an address after an unsubscribe or a bounce.
Anything that leaves your company waits. That's every email, LinkedIn message, SMS and marketplace proposal an agent writes. Deal stage changes and contact merges or deletes wait too. You'll find them all in one queue, grouped by agent, with the evidence the agent used laid out beside the draft.
SalesCrew's default approval policy
| Action | Default | Who acts |
|---|---|---|
| Research, enrichment, tagging, scoring, summaries | Auto | The agent, logged |
| Classify a reply, pause a cadence on reply, suppress on unsubscribe or bounce | Auto | The agent, logged |
| Any external send: email, LinkedIn message, SMS, marketplace proposal | Review | A person approves |
| Create a deal from a positive reply, book a meeting | Review, then auto after a clean week | A person, then the agent |
| Stage changes, contact merge or delete | Review | A person approves |
| Any spend: credits, connects, ads | Human only | A person, never an agent |
SalesCrew's seeded policy table, September 2026. Every row is editable in Agents, Guardrails.
Guardrails that apply even after you approve
Approval isn't the last check. At send time, SalesCrew re-checks suppression, so an address that opted out after the draft was written never gets the email. Each contact gets at least 3 days between touches. Every mailbox and every domain has a daily cap, and sends only go out inside your business-hours window.
If something looks wrong, you don't hunt through settings. The kill switch on the Agents page stops every agent with one control, and the audit log shows what each one did and when. Cadence steps you write and launch yourself still send on your schedule, inside the same caps.
Auto mode is a setting, so check who can change it
Questions
- Can I let a SalesCrew agent send some emails on its own?
- Yes, if you choose to. Each agent has four modes: off, draft, review and auto. In auto, the agent sends and logs the action, and anything below its confidence threshold drops back to review. External sends start in review on every instance, so auto is always your decision.
- How do I approve an AI email in SalesCrew?
- Open the approval queue. Items are grouped by agent, and each shows the draft, the thread and sources it used, its confidence and its cost. Approve it, edit it and then approve, or reject it with a reason. The reason feeds the agent's next run.
- Does the SalesCrew approval queue cover LinkedIn and proposals too?
- Yes. The default policy puts every external send in review: email, a LinkedIn message, SMS and a marketplace proposal such as an Upwork bid. Any spend, like credits, connects or ads, is human-only and never runs from an agent.
- Can an AI client connected over MCP skip the SalesCrew approval queue?
- No. An MCP tool has the same effect as its button in the app, and an external send from an AI client goes through the same queue by default. Every call is written to the audit log with the token that made it.