Autonomy with handbrake.
Each action of each agent is recorded with name, permission and time. Whatever you decide needs approval, ask for it. And everything an agent does can be reversed.
Waiting for approval · out of catalog price
Conversation · +52 55 1234 5678 · 11 min agoWhat an agent can do and what remains on the human side.
This is the default distribution. Everything can be adjusted by board, by calendar and by role, but nothing opens without you opening it.
| Action | Agent | team person |
|---|---|---|
| Reply in chat | Yes, within your catalog | Yeah |
| Create and edit contacts | Yeah | Yeah |
| Merge or delete contacts | No | Yeah |
| Create and move opportunities | Yes, by authorized board | Yeah |
| Mark won or lost | Yes, if the board allows it | Yeah |
| Schedule and reschedule appointments | Yes, within the configured schedule | Yeah |
| Create resources or change schedules | No | Yeah |
| Save form responses | Yeah | Yeah |
| Edit a closed answer | No | No |
| Save configuration changes | Proposes; does not save | Yes, upon approval |
They do the work. You remain in control.
You'll never wonder what happened while you weren't looking. Each action has an agent, time and record, and all can be undone.
Calendar Agent Roma branch availability changed
Calendar · Dr. RuizDeals Agent moved an opportunity to Submitted Proposal
Sales Dashboard · $2,400Chat Agent requested human support for an out-of-catalog price
Conversation · +52 55 1234 5678Forms Agent submitted the medical history form
Contact · Mariana CruzReminders Agent set a reminder 24 hours before
Appointment · tomorrow 5:00 pmContacts Agent created the profile and labeled it New lead
Contact · Mariana CruzWinning and losing is out of your reach: that decision is still yours.
Each agent declares what it can touch: "3 of 7 boards · writing on 2." Reading, writing and win/loss actions are granted separately.
Out-of-scope work does not run: the agent stops, notifies the owner with the reason and wait time, and waits for approval.
Turn off AI and write without leaving the thread. The agent steps aside and resumes when you reactivate it.
You test personality, playbooks and permissions without spending real messages or touching a record of your operation.
Each action is linked to the record that it modified. If something should not have happened, it is undone from the log itself.
One screen shows what's ready, what's broken and what's left to configure, with a suggested next step.
Autonomy, but with your hand on the brake.
Start with the agent requesting approval for everything, then expand permissions when its work earns your trust.
See all five modules