Microsoft is launching hosted agents in the Azure AI/Microsoft foundry Agent Service (public preview), providing per-session VM-isolated sandboxes with persistent filesystems designed specifically for production enterprise AI agents. Key features include scale-to-zero economics with filesystem-preserving resume, predictable cold starts, per-agent identity via Microsoft's own identity system (OBO), BYO VNet support, and built-in observability. The platform supports any agent framework (LangGraph, AutoGen, OpenAI SDK, Claude SDK, etc.) and any model, deployed via a single `azd deploy` command. Also announced alongside: a unified tool management layer (MCP-compatible), managed long-term memory, Microsoft Agent Framework v1.0 (merging AutoGen and SemanticKernel), and enterprise governance features including DLP and responsible AI guardrails.
