Blog Post
Bringing AI-Powered SRE to Your On-Premise Server Estate For Unifying Ops
Useful walkthrough, and the telemetry first rule is the part most people skip. Two thoughts from the small estate side:
- Arc, the Azure Monitor Agent, data collection rules and a Log Analytics workspace are a solid foundation, but for a team with a few dozen Windows machines, and workstations as well as servers, that is a lot to stand up before the agent can help. A short "minimum viable setup" section would help smaller teams decide if it is worth it.
- Human approval plus an audit trail for anything that acts is the right default, and I hope that stays the case as the agent gets more capable.
For readers whose estate is mostly Windows endpoints and who want something lighter, there are agent-based tools that take the same diagnose and fix approach without the Azure plumbing. TFalke is one example (disclosure: I work on it): an AI agent investigates a device and fixes the problem with a command that is shown for approval first. I would see it as a different fit for endpoint fleets, not a replacement for Arc and Azure Monitor on server estates.
WinEndpointOps - I understand the concern for large estate, but the logs has to be exported to Azure Log Analytics before SRE Agent can perform its investigation. While any enterprise customer onboards Azure ARC, they usually do via add multi server script, similarly once the servers are onboarded through powershell command and parsing the list of onboarded servers, customer can enable Azure Monitor Agent for Arc Enabled Servers, followed by configuration of DCR.
With regards to Human Approval Gates, yes thats the feature of Azure SRE Agent and required as part of enterprise architecture.
There could be other 3rd party / open source tools providing similar capabilities but this blog is focused on SRE Agent.