Microsoft’s Agent 365 Tries to Be the AI Bot Boss (www.wired.com)

Microsoft today launched Agent 365 in early access — a management layer for enterprise generative AI agents rather than a development platform. The product offers a centralized registry of active agents with unique IDs, per-agent configuration and permission controls, and real-time monitoring and scanning of agent activity. It can manage agents built with Microsoft tools or third-party systems, letting administrators audit usage, change access to business applications, and enforce security policies across an expanding “fleet” of automation that Microsoft executives say can scale to hundreds of thousands or millions of agents inside large organizations. For the AI/ML community this frames the next phase of productionizing agents: governance, observability and access control rather than model performance alone. Agent 365 addresses practical risks—eg, prompt-injection, runaway automation, and data-exfiltration—by adding auditing and runtime protections, but it doesn’t solve fundamental brittleness in agent behavior. The tool signals a commercial push to normalize wide agent deployment inside enterprises and raises questions about how well current defenses and governance practices will hold up as organizations adopt ever-larger populations of autonomous workflows.