There's a moment in almost every incident I've worked where the problem stops being technical. The fix might take five minutes. What takes forty is getting everyone onto the same page: who has access to the AWS account, who owns the downstream service, who's allowed to approve a restart, and which of the six open chat threads is the real one.
So, when I went through what Microsoft announced for Azure SRE Agent at Ignite 2026, I wasn't looking for "smarter AI." I was looking for whether it fixes that problem. And this time, mostly, it tries to.
I learned this on a project of my own. I built a food ordering site, and one day "Add to cart" just stopped working. The cause turned out to be a memory leak. Once we knew where to look, finding and fixing it was the easy part.
The hard part came before that. A new developer on the project had to work out how the app's logic fit together and which flow was actually broken, and that took far longer than the fix. The bug was simple. The context around it wasn't.
That's the gap I want tools like SRE Agent to close. Writing the fix was never the bottleneck. Understanding the system fast enough to know where to fix it was.
Quick context: SRE Agent went generally available in March 2026. Everything below is new at Ignite and, apart from one private preview, in public preview. Keep that in mind before you wire it into production on-call.
1. It joins the incident chat: Teams (and Slack)
This is the one I care about most.
SRE Agent can now be added to the Microsoft Teams chat where the incident is being run. You @mention it like a teammate. It pulls evidence from your connected systems and replies in the thread, so the investigation happens where people already are, not in a separate portal tab nobody has open.
The guardrails matter here too:
It runs within its assigned Azure RBAC permissions. It doesn't get extra access because it's an "AI".
Admins decide where it can take part and which actions need approval.
Anything that could touch production waits for an authorized human approver.
Slack support is in private preview, so Slack-first teams will have to wait a little.
My take: The best tooling disappears into your existing workflow. An agent that needs me to context-switch at 3 AM won't get used. One that answers in the war-room thread might.
2. Your multi-cloud reality is finally supported: AWS and Google Cloud connectors
Very few of us run "just Azure." Before this release, getting SRE Agent into AWS meant going through the AWS MCP server. Now there are built-in connectors for AWS and Google Cloud (public preview).
The design choice I like: they federate from an Azure managed identity, so there are no access keys to manage. Each cloud's own IAM still decides what the agent can see.
My take: No long-lived keys is the right call. Every static credential is a future incident, and I'd rather a new tool not add more.
3. Agents talking to agents (A2A)
This is the most forward-looking piece. SRE Agent now supports agent collaboration on the open A2A protocol (public preview).
In practice, your agent can ask another team's agent a question during an investigation and still own the incident. The other agent answers using its own tools and its own permissions, and its team approves who's allowed to call it.

Join the conversation! Your thoughts help the community grow.