0:00
50:34
Rewind 15 seconds
Fast Forward 15 seconds

This week Richard, David, and Cyrus spend less time gawping at shiny features and more time on the awkward bit that follows, which is who is supposed to own the behaviour, risk, and cost once agents start doing real work. David opens with Microsoft’s architecture guidance on agent design, arguing that testing AI systems needs a behaviour contract rather than blind faith in prompts or model choice.

Richard then picks up the thread from the enterprise side with GitHub Copilot Fusion and Agentic Workflows. The interesting part is not that GitHub can now orchestrate models more cleverly. It is that agentic work is becoming something you meter, govern, and supervise, a bit less like a helpful assistant, and a bit more like a workforce problem wearing a developer badge.

Cyrus brings the security weight, covering ISO 27090, the PHP remote code execution mess, new Intune baseline housekeeping, and passkeys for B2B users in Entra. The episode rounds out with Foundry Local, sovereign model pressure in Europe, Dataverse writeback through Copilot Cowork, and why Microsoft’s own Storm-0940 reporting should make any organisation look much harder at app permissions and service principals.

Links

Socials

Music

Null Invocation, Monochrome Pulse https://is.gd/b1kNU9

More episodes from "Cloudy with a Chance of Insights | A Microsoft Cloud (Azure, M365) Show for Architects & Engineers"