Agents operate, humans govern: Scale your operations and reduce toil with Azure SRE Agent
The New Stack TNS Staff
Microsoft says Azure SRE Agent can spot incidents, suggest fixes, and even draft PRs before humans join the bridge. The pitch is simple: less toil for SREs, and faster mitigation when production goes sideways.
Based on reporting by The New Stack, TNS Staff — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Microsoft is pushing Azure SRE Agent as a way to hand routine operations to software while keeping humans in charge of the final call. The idea is not just to answer alerts, but to pull in telemetry, deployment history, incident systems, and runbooks, then tell engineers what is likely broken and what to do next.
Sanchit Mehta, one of the head engineers behind the product, says the agent correlates blast radius, deployment changes, recent changes, and rollouts to narrow down the cause. In some cases it can even draft the pull request for the fix. That matters at 3 a.m., when a person trying to piece together a live incident is still bouncing between dashboards and trying not to guess wrong.
Microsoft says more than 3,000 of its service teams already use Azure SRE Agent for investigations, root cause analysis, incident response, code fixes, automatic mitigation, proactive detection, data analysis, and reporting. Inside Microsoft, the system has handled more than 1.8 million incidents, and many were mitigated in minutes. The company also uses it to improve itself, with custom agents for code review, deployment, evaluation, and monitoring.
The pitch gets stronger when Microsoft talks about what the agent can do before a human even notices the problem. Shamir Abdul Aziz, lead program manager for the product, says that for some internal teams more than half of incidents are handled autonomously because they fall into what he calls safe operations: a restart, a scale-out, a rollback, or a customer change-order request. The humans set the rules, the agent runs within them.
That governance layer is the real story here. Azure SRE Agent uses role-based access control, tool-access policies, hooks, verification, and telemetry so teams can see what the agent did and why. It connects to Azure Monitor, Application Insights, Log Analytics, Azure Resource Graph, Azure DevOps, GitHub, and outside systems through MCP connectors. And Microsoft is explicit that nobody should just switch it on and hand over the keys. The product now comes with a 30-day trial and no always-on charges, which is a pretty clear sign the company wants teams to start small, then let the machine earn more trust one incident at a time.
My take — AI-written commentary, not fact-checked reporting
This is the right bet. The fantasy that humans should babysit alerts forever is exactly the kind of expensive nostalgia that keeps SRE teams stuck in the mud. The useful part isn’t the agent’s confidence; it’s the guardrails, because nobody needs another black box with admin rights and a bad attitude.
Read more about this at: The New Stack