
Microsoft says organisations have registered more than 50 million AI agents in Agent 365 since the management platform launched in May. At Ignite 2026, the company is expanding the service with business analytics intended to show where agents are being used, what they cost and whether those deployments appear to deliver value.
The 50 million figure is striking, but its meaning needs care. Microsoft is counting registered agents, not necessarily agents used regularly or performing valuable work. Registration measures the scale of an emerging management problem. It does not measure productivity gains.
Microsoft’s Ignite announcement says Agent 365 Business Analytics will provide views of usage, spending forecasts and recommendations. That feature is in Frontier preview, rather than generally available. It is aimed at a question many executives now face: after purchasing and building AI agents, how do they tell which ones are useful enough to keep?
The company is also broadening governance beyond its own tools. It says agents built on platforms including Amazon Bedrock, Google, Salesforce, Databricks, Snowflake, Anthropic and OpenAI can receive identities through Microsoft Entra Agent ID, with support varying by platform and some capabilities still in preview. A separate risk view is designed to give IT and security teams a more consistent picture of agents across Microsoft services.
An agent can read documents, call services and sometimes act on behalf of a user. That makes ownership, permissions and lifecycle management more consequential than with a conventional chatbot. A company needs to know who created each agent, what it can reach and whether its access should survive when a project ends or an employee changes roles.
Microsoft is positioning Agent 365 as a common place to answer those questions. The strategy could be valuable in enterprises using several model providers, although its usefulness will depend on how completely it can observe and control agents outside Microsoft’s own ecosystem. Preview labels and platform-specific limits will matter in real deployments.
TechBooky has separately covered the new restrictions for Windows AI agents. Agent 365 addresses a related problem at organisational scale. The most important test is not whether another million agents can be registered, but whether administrators can tell which ones are safe, accountable and worth the expense.







