Microsoft just made deploying AI agents roughly as complicated as ordering coffee. The company’s Foundry platform, part of Azure AI Foundry, has moved its hosted agents to general availability, giving enterprises a managed runtime that works with essentially any framework, programming language, or AI model they want to throw at it.
The timing is deliberate. Showcased during Microsoft Build in early June 2026, hosted agents now offer sandboxed environments, auto-scaling, observability baked in, and the ability to deploy source code directly without wrestling with container builds.
What’s actually under the hood
The runtime supports Python 3.13 and 3.14, along with .NET 10, running inside hypervisor-isolated sandboxes. Each agent gets its own secure environment, completely walled off from everything else.
Availability spans more than 20 Azure regions, including East US 2, North Central US, and Japan East. Pricing follows a container-usage model, meaning companies pay for what they actually consume rather than reserving capacity they might not need. Observability comes through OpenTelemetry integration, giving teams real-time metrics on how their agents are performing.
