Letting AI operate production Azure — what it gets wrong

Real failure stories from AI-assisted Azure operations, each with the guardrail that fixed it, and the patterns that worked.
Abstract
Real failure stories from AI-assisted operations, each with the guardrail that fixed it:
- an Azure GET can’t prove a property is settable (recommended four times off an empty read);
- advising a change from a stale read in the middle of a cutover;
- restating a verified fact until it’s wrong;
- a disk caching change that hot-detached a production disk;
- “a freeze is not a control”.
Plus the patterns that worked: subagents, cross-model review with gh copilot, and grep-verifying every HIGH finding. It’s honest, funny, and very MVP.
Materials
Also at this event
Agents need an identity, and your IT department will say no
A practitioner's account of giving AI agents their own identity: where it breaks down, how an AI gateway bridges Entra and legacy backends, and a governance model that holds up.
Speaker Talk Dublin Slides
Around this page
Explore connections. Hover to highlight neighbours; click or use Enter to visit. On touch, tap to highlight and tap again to visit. The drawing shows up to 40 pages and topics (10 on small screens); all connections remain in the list.
All local connections
- Letting AI operate production Azure — what it gets wrong talks
- Infrastructure as Code categories
- MVP tags
- AI and Agents categories
- Site Reliability Engineering tags




