Runbook-guided mitigation that waits for approval
An SRE agent investigates an Azure incident, proposes a mitigation from your runbooks, and acts only after an admin approves.
The use case
Teams want agents to help fix incidents, not just describe them, but cannot risk unreviewed production changes. Azure SRE Agent investigates alerts across Azure resources, uses attached runbook context to propose a mitigation such as a restart or scale-out, and in review mode waits for an administrator to approve each action. The whole investigation stays in one thread with the evidence. Teams can loosen this for low-risk actions later. A person still approves anything that touches production, and the team reviews the agent's actions after each incident.
Example workflow
- Connect the agent to subscriptions and attach runbooks
- Set the run mode to review so actions need approval
- Agent investigates alerts and proposes a mitigation
- Admin reviews the evidence and approves or rejects
- Review agent actions in the post-incident meeting
Tools
- Azure SRE Agent
- Azure Monitor
Risks and guardrails
- Keep approval required for production changes
- Runbooks that are out of date lead to bad suggestions
- Scope agent permissions to the resources it needs
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Microsoft Learn: Azure SRE Agent overview: learn.microsoft.com/en-us/azure/sre-agent/overview