Autonomy Level Dashboard
Autonomy is a governed property of each use case, not a prompt setting. Elevated-risk actions — refunds, plan changes — trigger a human approval step at the gateway before any downstream dispatch.
The six-level scale
Live population of the estate across levels
Reads and answers. No side effects of any kind.
Proposes an action for a human to take elsewhere.
Drafts and stages the action; a human commits it.
Executes reversible, single-subject, rate-limited actions.
Executes material actions behind a mandatory approval step.
Executes without per-action approval. Not granted in this estate.
Use cases by autonomy
Each agent's granted level and its risk classification
| Use case | Agent | Level | Risk | Approval gate |
|---|---|---|---|---|
| Tier-1 service deflection and invoice questions | Salesforce Agentforce — Service Copilot | L4 execute-minimal | moderate risk | human approval |
| Employee self-service lookups inside Teams | Microsoft Copilot — Workplace Agent | L2 recommend | moderate risk | not required |
| Billing dispute resolution with refund authority | Custom Care Orchestrator | L5 execute-controlled | high risk | human approval |
| Autonomous first-line network remediation | OpenShift NetOps Agent | L4 execute-minimal | high risk | not required |
| Revenue assurance and reconciliation preparation | Internal Finance Analyst Agent | L3 prepare | critical risk | human approval |
| Runbook execution for recurring incidents | Internal Ops Runbook Agent | L4 execute-minimal | moderate risk | not required |
| Unreleased prototype under evaluation | Sandbox Prototype Agent | L1 answer-only | low risk | not required |
Pending human approvals
Elevated-risk actions held at the gateway
No actions awaiting approval
Trigger an elevated-risk action to see the gate fire.
Call trail
Select any call in the live stream to open its full trail
No call selected
Click a row in the live MCP flow to inspect identity, policy decision, token and result.