Microsoft says its Smart DRI Agent is now used by more than 40 teams to help engineers investigate production incidents, bringing telemetry, troubleshooting guidance and similar past cases into Teams. The company estimates it saves 600 engineering hours a month, though that figure is its own assessment, not an independently measured result.
Microsoft AI Watch analysis
What happened
The agent gathers relevant traces, logs and exceptions from the minutes before an incident, searches troubleshooting guides and surfaces context from similar issues. It can also send proactive incident notifications and morning summaries, rather than waiting for an engineer to ask the right question in the right portal.
Microsoft says the system has handled more than 7,800 AI-assisted interactions and sent nearly 6,000 proactive notifications. The company describes it as a tool for designated responsible individuals, or DRIs, who take turns monitoring services and coordinating responses. It is built with Microsoft Semantic Kernel and Azure AI Search, with custom logic added after the team found search alone did not consistently return useful matches.
Why it matters
Incident response often begins with a scavenger hunt across dashboards, documents and people who know where the useful information lives. Putting that context beside the engineer in Teams could reduce the delay before someone can act, particularly when a DRI is new to a service or covering an unfamiliar system.
The agent currently recommends and organises information; Microsoft says a future version may take steps to mitigate incidents, with the human DRI kept in the loop to approve them. That distinction matters. A useful briefing can save time, but an automated intervention in a live service needs a rather different report card.
Our read
The practical lesson is less “add an agent” and more “connect it to the operational material people already need, then test whether it finds the right thing”. Microsoft says its team added custom similarity logic and configurable workflows, and stresses that the agent depends on an established live-site reliability culture. No chatbot can compensate for missing ownership or a service nobody is monitoring. A surprisingly old-fashioned requirement, but there we are.
What to watch
- Whether Microsoft reports measured incident-resolution times, alongside its estimated hours saved.
- How often the agent’s suggested troubleshooting steps and similar-incident matches prove useful.
- Whether future incident mitigation keeps clear human approval and accountability.
Discussion spark: Would you let an AI agent take approved mitigation steps during a live incident, or should it stay strictly in the role of briefing the human responder?
Sources and evidence
- Inside Track – Automating engineering operations at Microsoft with a Smart DRI Agent – Microsoft (15 October 2026, 15:45 UTC)
not affiliated with or endorsed by Microsoft