Three trust levels
Observer
The Manager monitors and logs but takes no action. This is the default starting level.- Detects issues and anomalies
- Records observations in the activity log
- No user notifications beyond the log
Advisor
The Manager detects issues, proposes specific fixes, and asks for your confirmation before acting.- Sends notifications with proposed actions
- Waits for explicit approval before executing
- Tracks acceptance rate to build trust score
Autonomous
The Manager executes fixes automatically without requiring confirmation.- Acts immediately on detected issues
- Logs all actions for audit trail
- Can be reverted to Advisor at any time
Why this matters
Autonomous infrastructure agents need guardrails. Giving an AI full control from day one is risky. The trust progression model:- Builds confidence — You see the Manager’s judgment before trusting it
- Prevents damage — Bad suggestions are caught before execution
- Creates audit trails — Every action is logged regardless of trust level
- Enables rollback — You can reduce trust level at any time
Further reading
Guard setup
Configure Guard’s trust level
Architecture
How Managers fit into the overall system