Rediger

Del via


Run modes in Azure SRE Agent

Run modes control whether your agent asks for approval before taking actions or acts on its own. The key distinction is between Azure infrastructure operations (which have system-enforced approval) and other actions.

Note

Run modes control the approval workflow, deciding whether the agent should ask before acting. Permissions control resource access, determining whether the agent can reach a resource. The agent needs both conditions satisfied to act.

Two modes

Screenshot of a comparison of review mode where the agent proposes and the user decides versus autonomous mode where the agent executes immediately.

The following table summarizes the two available modes:

Mode What happens Best for
Review Agent proposes an action; you approve or deny Production systems, critical infrastructure
Autonomous Agent executes immediately and reports what it did Nonproduction environments and trusted recurring tasks

Review mode

Review is the default mode. Your agent investigates, identifies a fix, and asks for your approval before executing Azure infrastructure operations (Azure CLI commands, Azure Resource Manager operations, and similar write actions).

Note

Review mode shows Approve and Deny buttons only for Azure infrastructure operations. Other actions, like sending emails, posting to Teams, or querying external data sources, proceed based on the agent's available tools and permissions.

The following example shows what you see for Azure infrastructure actions:

I found that app-service-prod is running slowly due to high memory usage.

Proposed action: Restart App Service 'app-service-prod'
This may cause brief downtime (30-60 seconds).

[Approve] [Deny]

Select Approve to execute the proposed action, or Deny to stop it. Only SRE Agent Administrators can approve actions.

Autonomous mode

In autonomous mode, your agent investigates and executes actions without waiting for approval. Use this mode when you trust the agent to handle the situation.

The following example shows what you see in autonomous mode:

I found app-service-staging was running slowly.

Done: I've restarted app-service-staging. Memory usage is now normal.

Where to configure run modes

Set run modes per response plan and per scheduled task. You don't set run modes at the agent level.

Automation type Default mode Options
Incident response plan Autonomous Review, Autonomous
Scheduled task Autonomous Review, Autonomous

Set the Agent autonomy level when you create or edit a response plan or task:

Edit incident trigger dialog showing Agent autonomy level with Autonomous and Review options.

Agent-level default

Settings > Basics shows the agent's global mode. Set this mode when you create the agent (defaults to Review). It serves as the fallback when no per-response-plan or per-task mode is set.

Recommendations

The following table provides guidance for choosing the right mode based on your scenario:

Scenario Recommended mode
Production incidents Review
Staging/dev incidents Autonomous
Daily health checks Autonomous
Cost and usage reports Autonomous
Security alerts Review

Start with review mode. Observe what the agent recommends for two to four weeks. When you find patterns you consistently approve, switch those specific triggers to Autonomous.

How permissions interact with run modes

The agent behaves differently depending on the assigned permissions, the execution mode, and the type of action it attempts to make. In all cases, if the agent doesn't have the required permissions, it requests temporary access through Microsoft Entra OBO flow.

Read-only actions

The following table details how the agent behaves when it attempts a read-only operation that requires elevated permissions.

Agent has permission? Execution mode Agent behavior
Yes Review Uses its permissions to perform the action
No Review Prompts for temporary access to perform the action on behalf of the user
Yes Autonomous Uses its permissions to perform the action.
No Autonomous Prompts for temporary access to perform the action on behalf of the user.

Write actions

The following table explains how the agent behaves when it tries to perform a write operation.

Agent has permission? Execution mode Agent behavior
Yes Review Prompts for consent to take action, and then uses its permissions to perform the action once consent is granted.
No Review Prompts for consent to take action, and then prompts for temporary access to perform the action on behalf of the user.
Yes Autonomous Uses its permissions to perform the action.
No Autonomous Prompts for temporary access to perform the action on behalf of the user.