Microsoft opened Ignite 2025 on November 18 by introducing Agent 365, a control plane for deploying, securing, monitoring and governing AI agents across an organization. Microsoft framed it as the management half of a larger agent platform: teams build agents in Microsoft Foundry, the new name for Azure AI Foundry, manage them through Agent 365 and ground them in company data through new "IQ" context layers. For developers, the announcement signaled that enterprise agents would increasingly be treated like users and applications, with an inventory, access policies and oversight.

Key Facts#

  • Event: Microsoft Ignite ran from November 18 to 21, 2025, and Agent 365 was one of its headline announcements.
  • What Agent 365 is: a central place to secure, monitor and govern the agents an organization builds or uses.
  • Initial availability: early access through Microsoft's Frontier program.
  • Foundry rebrand: Azure AI Foundry became Microsoft Foundry.
  • More models: Microsoft added Anthropic's Claude models to Microsoft Foundry and said it was the only cloud offering both Claude and GPT frontier models on one platform.
  • Context layers: Work IQ, Fabric IQ and Foundry IQ were presented as the ways to give agents business context.
  • General availability: Agent 365 became generally available on May 1, 2026.

What Happened#

Microsoft's Ignite messaging described a complete lifecycle for AI agents, from idea to deployment. On the building side, Microsoft Foundry is the platform for creating agents and choosing models, now including Anthropic's Claude family alongside OpenAI's GPT models. On the context side, Work IQ, Fabric IQ and Foundry IQ connect agents to the data in Microsoft 365, Microsoft Fabric and Foundry's knowledge services. On the operations side sits Agent 365.

Agent 365 targets a problem that grows with every successful pilot: once teams can create agents easily, organizations quickly lose track of how many exist, who owns them, what they can access and what they are doing. Coverage of the launch described Agent 365 as the central control system for managing agents inside an organization, giving administrators one place to secure, monitor and govern every agent they build or use. Microsoft made it available through its Frontier program, which offers early access to new AI capabilities.

Microsoft's training material for the product, written after general availability, shows how that governance works in practice. Administrators work in the Microsoft 365 admin center, where agents are registered through an approval workflow, access controls are applied according to organizational policy, and a central view shows active agents, their approvers and the data they reach. Activity data can then be reviewed to enforce governance across what Microsoft calls the agent estate.

The context layers matter just as much for developers. Foundry IQ, for example, is documented as a managed knowledge layer that turns enterprise data into reusable, permission-aware knowledge bases for agents. It connects to sources such as Azure Blob Storage, SharePoint, OneLake and the web, uses an agentic retrieval pipeline that splits complex questions into parallel subqueries, and runs on Azure AI Search.

Background#

By late 2025, Microsoft had assembled most of the pieces of an enterprise agent stack. Microsoft.Extensions.AI reached general availability in May 2025, Microsoft Agent Framework entered preview on October 1, 2025, and the MCP C# SDK had been available since April. Microsoft's Ignite theme centered on what it calls the Frontier Firm, an organization that uses Copilot and agents throughout its work.

Governance was the missing layer. Security teams already manage users with identity systems, devices with endpoint management and data with compliance tools, but autonomous agents that act on behalf of people do not fit neatly into any of those. Agent 365 was Microsoft's attempt to create that category, and adding Claude models to Foundry at the same event reinforced that the platform would be multi-model rather than tied to a single provider.

Why It Matters for Developers#

The practical impact is that agents built for enterprise customers will increasingly be expected to behave like well-governed software, not experiments. That changes how you design them:

  • Give agents their own identity. Microsoft Entra Agent ID, which reached general availability in May 2026, provides purpose-built identities and OAuth flows for agents. Design agents to run with their own least-privilege identity rather than borrowing a user's or an application's broad permissions. Our comparison of identity platforms covers the underlying Entra concepts.
  • Plan for registration and approval. If agents must pass an approval workflow before use, they need clear ownership, descriptions of the data they access and a documented purpose. Build that metadata into your delivery process.
  • Instrument everything. Governance depends on activity data. Emit OpenTelemetry traces for agent runs and tool calls, as described in our guide to observability and cost control for LLM apps.
  • Use Foundry deliberately. Microsoft Foundry is now the umbrella for models, agents and knowledge. .NET developers can work with it through the Foundry integration in Microsoft Agent Framework or through IChatClient, as explained in Azure OpenAI and Microsoft Foundry for .NET.
  • Compare managed retrieval with your own. Foundry IQ offers permission-aware knowledge bases and agentic retrieval as a service. Weigh that against a custom pipeline using the techniques in our RAG in .NET guide, considering control, cost and data residency.

The multi-model angle also matters. With Claude and GPT models side by side in Foundry, choosing a model becomes a per-workload decision. Code that depends on abstractions rather than a single vendor SDK can take advantage of that choice.

What's Next#

Agent 365 moved from Frontier early access to general availability on May 1, 2026, and Microsoft Entra Agent ID also reached general availability that month, giving the governance and identity layers production status at roughly the same time. Foundry, meanwhile, has continued to add models from multiple providers, including Microsoft's own MAI models in 2026. The open questions are how well Agent 365 handles agents built outside Microsoft's platforms, how licensing will scale as organizations deploy hundreds or thousands of agents, and how the IQ layers will balance convenience against the control that regulated industries require.

Sources#