Review policy suggestions

Settings → AI Configuration → Learning is the review queue for suggested changes to your tool, budget, and checks-gate policies. It needs the view_policy_learning permission.

The loop is deliberately one directional: something observes how work actually ran, proposes a change, and a person decides. Nothing on this page changes a policy without you.

Where suggestions come from

A suggestion is derived from observability signals over a trailing window: which tools were used, what runs cost, where gates fired. Those signals come from the platform’s own telemetry pipeline.

A new workspace has nothing to suggest yet

Suggestions are derived from a trailing window of your own activity, so a workspace that has not run much yet shows an empty list. The review, apply, and rollback controls all work; there is simply nothing for them to act on until enough has happened to draw a conclusion from.

Read one

Each suggestion carries:

• The change: which policy, and what it would become.
• The rationale: the observation behind it, in a sentence.
• A confidence: how strongly the signals support it.

Read the rationale before the change. A suggestion to raise a tool budget because a class of run kept stopping short is a different decision from the same change proposed because one run was unusual.

Decide

• Apply writes the change to the policy and records the previous value.
• Reject declines it, with a reason you type. The reason is worth writing, because it is what a later reviewer sees when a similar suggestion returns.
• Rollback restores the previous value of a suggestion you applied. It is available on applied suggestions only, which is what makes applying a reversible act rather than a commitment.

Every decision is recorded in the audit log under the person who made it.

A working order

1. Leave suggestions alone for a week rather than acting on the first one. A single run is not a pattern.
2. Apply one at a time. Two applied together make the next week’s signals unreadable.
3. Watch the area the change touches for a few days.
4. Roll back rather than tuning around a change that made things worse.

Next steps

• Budgets: the policy most suggestions touch.
• The audit log: where applied and rolled-back changes are recorded.
• Router policy: the other page whose controls are not yet enforced.