Shipping Production AI Agents on Azure
summary
Velo is an AI-powered DORA metrics platform running ASP.NET Core 9, Angular 19, Azure Container Apps and Azure SQL Serverless with Microsoft Foundry models (GPT‑4o / 5.2 / 5.4 mini) used on demand. Tenancy maps to the Azure DevOps organization ID and auto‑provisions on first webhook, metrics compute incrementally and tolerate duplicates/out‑of‑order events. An agent layer sits beside the data path so metrics remain correct; a production install shows change failure rate at 66.7% and MTTR 8.2h. Identity role gaps cause silent 403s; four RBAC settings fix it.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:04Production AI agents differences
- 2:10Velo architecture
- 3:36DORA metrics constraints
- 5:49Stateless compute isolation
- 6:09Handling hostile inputs
- 7:37Tenant auto-provisioning
- 8:42Agent layer design
- 9:45Live demo Velo dashboard
- 12:40Debugging identity failures
- 15:12Prototype vs production
- 16:37Migrating to new frameworks
- 17:38Cost and scale
- 18:41Five lessons deployment
- 19:24Summary
related talks
16mLeveraging Azure AI Foundry to Build Intelligent Release Agents
05h 17mBuilding AI Applications
52mDesigning Production-Ready Agentic Systems: Lessons from the Field
43mUsing AI with App Service to deploy differentiated web apps and APIs
34mThe Hard Truth About AI Agents in Production
26mShip faster. Stress less. Idea to ops
40mOptimizing AI Applications for Production with Observability
40mBuilding secure and efficient AI solutions in Azure with Partners