brainis
Autonomy & Approvals

The approvals queue

Where AI proposals wait for a human: what arrives there, how to review efficiently, and how decisions teach the system.

1 min read

Any action an AI wants to take but may not execute alone becomes a proposal and waits in the approvals queue.

What lands in the queue

  • Actions above the scope's autonomy level or risk ceiling.
  • Verbs marked approval-always in the registry (destructive and high-impact operations).
  • Anything exceeding a monetary ceiling or attempted during quiet hours.
  • Bulk operations over large record counts, which stage for review rather than executing at once.

Reviewing a proposal

Each proposal shows what would change - before and after, on which records - plus the reasoning, the source (which agent or play), and a preview where the action supports it. Your options: approve, reject, or open the full run replay first. Approvals execute immediately and are logged with your name.

Reviewing efficiently

1
Handle proposals in batches at set times rather than as interrupts; Today shows the waiting count.
2
Use rejection deliberately - a rejected proposal with a corrected instruction improves the agent; a silently ignored one teaches nothing.
3
Watch for approval fatigue: if you approve the same verb from the same agent unchanged for weeks, that verb is a candidate for delegation in its contract.

Important: Approval permissions follow module roles - a proposal to change a deal needs approval from someone with the right level in Revenue OS, and contracts can name approver roles explicitly.

Still need help?

Our support team is available around the clock.

Contact Support