An alert tells you something is wrong. It does not tell you which evidence to trust, whether a recent change caused the problem, or whether the proposed fix will restore service.

This session follows an incident in an isolated Azure application using Azure SRE Agent. We’ll examine how the agent gathers operational context, develops a diagnosis, and proposes a mitigation—and how an engineer evaluates those recommendations before allowing a change.

Through a worked demonstration, we’ll distinguish symptoms from causes, inspect the evidence supporting the agent’s conclusions, and explore how Azure permissions and approval controls shape its ability to act. After a mitigation, we’ll verify application behaviour independently rather than treating the agent’s “resolved” message as proof of recovery.

We’ll also examine the preparation behind the demonstration: telemetry, resource access, response instructions, and any supplied runbooks. The aim is an honest assessment of AI-assisted operations, including where deterministic automation remains the better choice.

Attendees will leave able to:
- Assess whether an agent’s diagnosis is supported by operational evidence.
- Distinguish resource permissions from action-approval controls.
- Verify recovery and recognise when a mitigation has not removed the underlying cause.

Audience: Azure engineers, platform engineers, SREs, cloud architects, and technical operations teams.
Level: Intermediate.
Duration: 45 minutes; expandable to 60 minutes with deeper investigation and discussion.
Prerequisites: Familiarity with Azure resources, monitoring, and RBAC. No machine-learning knowledge required.
Format: Architectural explanation and a live demonstration, with a recorded fallback.
Session status: In development. The final demonstration and product-specific claims will be validated before submission or delivery.