Agent drafts
The agent prepares the send — personalized from your approved templates, with the lead's researched context attached.
Human-in-the-loop AI SDR
Nothing sends without your sign-off
The most common objection to AI SDRs is some version of: "I want to see the message before I trust the send." That's not a feature request — it's the correct default posture for anything going out under your name. Agent Salesforce is built around it: agents do the work, humans hold the pen.
The objection
"Show me the prompt before I trust the send"
Fair. Autonomous outreach that you can't inspect is how lists get burned and domains get torched — "AI slop" complaints across the category trace back to sends nobody reviewed. The answer isn't a promise that the model writes well. It's a structural guarantee: an approval checkpoint the agent cannot route around.
In Agent Salesforce, approval isn't a mode you toggle off and forget. It's a queue the agent must pass through for anything you haven't explicitly pre-authorized — with the full message, the lead's context, and the sequence step attached, so a review takes seconds, not a meeting.
The mechanism
How the approval queue works
Send pauses
New sequences and sensitive sends land in your approval queue — routed to Slack or email, full message and context included.
You decide
One tap approves, edits, or stops. Edits feed back into the sequence so the next draft starts from your version.
Decision logged
Every approval, edit, and stop is recorded with who decided and when — part of the same activity log as the sends themselves.
Progression
Review everything first. Loosen the reins when it's earned.
Most teams start in per-send review: every message pauses. As the drafts prove themselves, you pre-authorize more — a sequence that's run clean for weeks doesn't need per-message review, while a brand-new sequence or a strategic account always does. The dial is yours, per sequence and per segment, and every position of it is logged either way.
The result is autonomy you can defend: to your prospects, because nothing sloppy goes out; to your leadership, because every send traces to a named approval; and to yourself, because you saw the message before you trusted the send.
The record
Approval is only half of auditable
Every decision — approve, edit, stop — lands in the same structured activity log as every send, reply, and booking. Lead, sequence step, outcome, approver, timestamp. Synced to your CRM, replayable on demand. Governance isn't a promise; it's a query.
Go deeper
See it against the alternatives
See the queue before you trust the send.
Bring a sequence — we'll run it through the approval flow live, on your own messaging, in the first call.