Building an Operations Assistant on Azure That Requires Human Approval

adnan_hashmi · x · 2026-08-05

Building an AI assistant grounded in runbooks is the easy part; the real engineering challenge lies in the boundary between what a model proposes and what a human actually authorizes.

The author shares a reference app running on Azure that retrieves Markdown runbooks via Azure AI Search and calls Azure OpenAI. To ensure safety, the assistant can call deterministic read tools directly, but for control-plane actions like opening an incident, it records a pending action and returns a preview. A separate endpoint executes the action only after authenticated user approval.

Original post →

More from coding & agent

coding & agent channel →