Skip to content

What is the Azure SRE Agent?

Azure SRE Agent is an AI teammate for operating production on Azure. It watches your live estate, investigates incidents like an experienced site reliability engineer, explains the likely root cause with evidence, and, with your approval or within guardrails you set, remediates. It reasons across every tier: containers (AKS), the API front door (APIM), databases, caches, and the telemetry that ties them together, not just one service.

What it does

  • Detects & triages: correlates metrics, logs, traces, resource health and recent changes (Azure Activity Log + GitHub) to find what broke and when.
  • Investigates read-only: produces an evidence-backed root-cause hypothesis before anyone touches production.
  • Remediates safely: proposes a least-disruptive plan you approve, or acts autonomously within a bounded, RBAC-scoped posture.
  • Learns your operations: grounds in your runbooks and architecture, and remembers past incidents so recurrences resolve faster.
  • Extends to your team: custom specialist subagents, reusable skills, alert-triggered response plans, scheduled health checks, and a Python sandbox for reports.

Why it matters

  • Faster recovery (lower MTTR): tireless, parallel investigation shrinks the time from alert to root cause.
  • Less on-call toil: routine triage and well-understood fixes are handled for you, so engineers focus on the hard problems.
  • Governed by design: every action runs under Azure RBAC and is attributable and auditable; you decide how much autonomy to grant.
  • Consistent & institutional: the same sanctioned runbook steps run every time, and hard-won knowledge doesn't leave with the on-call.

You'll build exactly this across the hack

Onboard it read-only, investigate an incident, recover under approval, ground it in your runbooks, engineer a subagent + skill, let it recover autonomously within guardrails, command a major incident, and brief leadership.

How you control it: two independent dials

Permissions (Azure RBAC, what the agent's identity is allowed to do) and run mode (how it acts: Review proposes and waits for approval; Autonomous acts within its permissions and guardrails) are the two controls that matter all day. You start read-only and earn autonomy.

Control Options What it decides
Permissions (Azure RBAC) Reader → narrow scoped write What the agent's identity is allowed to do
Run mode Review · Autonomous How it acts: propose-and-wait vs act within guardrails

Everything else, including skills vs subagents, the knowledge base, and the agent's own telemetry resources, is in The rest of the picture below. Read it once when you have a spare minute; you do not need it all before you start.

Documentation you will return to all day: Overview · Permissions & run modes

The rest of the picture

Read this once when convenient. The two dials above are the core; everything here rounds out the picture.

  • Approved writes run as the agent's identity, not yours, so the Activity Log attributes them to the agent.
  • On-behalf-of (OBO), where an action runs under your credentials, is not available in this environment. That is precisely why scoping the agent's own identity matters.
  • Subagents are invoked explicitly, in natural language, by naming the one you want ("delegate this to your aks-triage subagent"). There is no slash-command syntax, and asking the agent to act as a specialist does not invoke one, it just adopts the persona.
  • Skills are matched against their description rather than guaranteed to fire. Name the skill when you want it used. Skills are capped at 5 concurrent active at a time.
  • Only a user with the Azure SRE Agent Administrator role can approve proposed actions.
  • The knowledge base accepts Markdown/text files (this repo's knowledge/), which is how you ground the agent in Aetherion's runbooks and standards.
  • Creating an agent also provisions its own Application Insights, Log Analytics, and managed identity, separate from the application's telemetry.

Documentation you will return to all day: Overview · Create and set up · Complete your setup · Permissions & run modes · Security overview · Team onboarding & memory · Subagents & extensibility · Skills · Agent hooks · Pricing & billing