Human-in-the-Loop Approvals for AI Agents: Why Sending Waits for a Manager
technology
How approval rules keep people in charge of external actions, and how governed learning keeps agent behavior reviewable.
Autonomy is not the goal. Dependable work is. Human-in-the-loop approvals let AI agents do the preparation while people keep the final say on actions that matter.
What an approval rule does
A rule defines which actions need sign-off before they happen. Coming soon: a procurement agent drafts renewal terms from the contract and the rate card, and sending anything outside the company waits for a manager to approve, by policy.
Which actions to gate
- Anything sent outside the company
- Changes to shared records
- Actions involving money or contracts
Low-risk drafting can move freely, so approvals stay meaningful instead of becoming noise.
Governed learning
Approvals are not the only checkpoint. When someone corrects an answer or rejects a draft, the correction is saved with who made it and why. An owner or admin reviews it before it changes how agents behave.
Rollback
If a change turns out wrong, you can roll it back.
A full activity record
A full activity record is also coming soon, so you can see what agents did and who approved it.
See Constella for enterprise for more.