Confidence Scoring & Escalation

Every agent result includes a confidence score, which expresses how likely it is that the output is correct and useful.

How Confidence Is Used

• Above threshold – Results may auto-complete without human review.
• Below threshold – Tasks are routed to human reviewers.
• Borderline cases – Workspaces can define custom behavior (e.g., always review for certain task types).

Workspace Thresholds

Each workspace can define a minimum confidence threshold:

• More conservative teams may require higher confidence for auto-approval.
• Teams experimenting with automation may set lower thresholds to increase throughput.

Escalation

Low confidence combined with:

• A critical priority task, or
• A history of similar failures

can trigger escalations such as immediate review or assignment to specific reviewer groups.

Confidence scoring helps balance risk and speed in a way that fits each team’s tolerance.