BRK420 Breakout Tue 17 Nov 16:30 PST
Deploy and run Microsoft Foundry models and agents on AKS
Brendan Burns, Jorge Palma
Session description
Microsoft's catalogue description
This session walks the integrated path: pick a model in Foundry and deploy it to your own AKS cluster, set a runtime tuning profile, and run it on GPU node pools with workload identity, policy, and built-in observability. It also covers Foundry hosted agents on AKS in both bring-your-own-cluster and managed modes, so you choose how much of the infrastructure you own. Leave knowing how to take a model or agent from Foundry to a governed, scalable production deployment on AKS.
Demo timestamps
Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.
- Skip to 0:05:20
AKS AI manager overview video
Short recorded intro to AKS AI manager, described as connecting the Foundry managed service to Azure Kubernetes Service, from model-as-a-service through to open-weight models on the customer's own infrastructure.
- Skip to 0:11:06
Caldova production-line application
Operations view of multiple production lines with an issue on line 7. Signals, readiness checks and equipment-manual links show the existing slow root-cause analysis path before agents and models are applied.
- Skip to 0:19:44
Plant-side nodes managed by AKS with Entra-authenticated models
Models are deployed onto cores running in the plants rather than Azure, with nodes managed by AKS. Models are accessed with short-lived Microsoft Entra ID tokens instead of API keys.
- Skip to 0:27:07
Agent Grid policies in the Azure portal
Agent Grid shows allow-lists for maintenance, telemetry and explicit APIs. The same blocked call seen in Foundry appears as a red line in the portal graph of agent API calls.