What is agent mode?

The dial that decides whether an AI agent only drafts, waits for approval, or acts and logs it.

An admin approves every new account by hand. Nothing is created until then. We reply by email; no newsletter, no sequence.

app.salescrew.io/inbox
The unified reply inbox with classified threads

Agent modeAgent mode is the setting that decides how much an AI agent may do on its own: off, draft (prepare only), review (queue for approval) or auto (execute and log). It lets a team raise autonomy one step at a time.

Why it matters

The same AI agent can be safe or reckless depending on one setting. In off mode it does nothing. In draft mode it prepares an email or a record update but never sends or saves it. In review mode the same output goes into an approval queue for a person to check. In auto mode it executes on its own and writes what it did to a log.

A team that skips straight to auto is trusting an untested agent with real actions on day one. The safer path runs each action type through draft and review first, watches a couple of weeks of outcomes, then moves that specific action to auto while leaving riskier ones in review. A drafting task, like summarising a call, can graduate faster than a sending task, like an outbound email a prospect will actually read. Agent mode is set per action type, not once for the whole system, so a team can trust an agent with scheduling while still reviewing every send.

How agent mode is used

  1. 1

    Start in off or draft

    The agent prepares nothing, or prepares output nobody sees acted on.

  2. 2

    Move to review

    Prepared actions land in an approval queue for a person to check before they run.

  3. 3

    Watch a clean period

    Track how often review catches a real problem versus rubber-stamps the output.

  4. 4

    Promote to auto per action

    Low-risk, well-tested actions move to auto while riskier ones stay in review.

  5. 5

    Keep a kill switch nearby

    Any mode can be reversed in one step if an agent starts misbehaving.

The mistake to watch for

Starting in auto to "see what it does". Start in review; move to auto per action class after a clean period.

Questions

How is agent mode different from a feature flag?
A feature flag turns a feature on or off for everyone. Agent mode is graduated per action type and controls how much independent judgment an agent gets, from none to full execution.
Does agent mode apply to the whole system or one action?
It applies per action type. A team can run drafting in auto while keeping sending in review, so autonomy grows where it has been earned rather than everywhere at once.
Who should be able to change agent mode?
Typically an admin or team lead, not the agent itself. Changing mode is a policy decision, and it should leave a record in the audit log like any other change.