Key Takeaways

  • Microsoft Agent 365 is now generally available for commercial customers
  • It provides discovery, inventory, and governance for AI agents across Microsoft, AWS, and Google Cloud
  • IT teams can use Intune policies to block Claude Code (OpenClaw) on managed devices via the Microsoft 365 admin center
  • Asset context mapping will show devices, MCP servers, identities, and cloud resources for each discovered agent
  • JPMorgan Chase reclassified AI investments from experimental R&D to core infrastructure with a $19.8B 2026 technology budget

The Facts

Microsoft has moved Agent 365 from its Frontier early-access programme to general availability for commercial customers, marking a significant milestone in the enterprise AI governance market. The product provides a unified inventory, observability, and control plane for AI agents running across an organisation's environment — whether they run locally on employee devices, as SaaS agents, or as cloud-deployed agents on AWS Bedrock, Google Cloud, or Azure.

The general availability announcement included a feature that will be particularly notable for IT administrators managing environments where employees have adopted AI coding assistants independently: the ability to use Intune policies to discover and block Claude Code (marketed internally as OpenClaw) on managed devices, accessible through the Microsoft 365 admin center.

This capability reflects the "shadow AI" challenge that enterprise IT teams are increasingly confronting. Employees are adopting AI tools — coding assistants, autonomous agents, research tools — faster than IT teams can evaluate, procure, and govern them. Agent 365's Shadow AI page in the M365 admin center gives IT teams visibility into which local agents are running on managed devices before deciding whether to block or permit them.

Multi-cloud governance is also a significant feature. AWS Bedrock and Google Cloud connections are now in preview, enabling IT teams to discover and inventory cloud agents deployed on other providers' infrastructure through a single Microsoft management interface. This addresses the agent sprawl challenge that intensifies as developers build agents using multiple AI platforms simultaneously.

JPMorgan Chase's concurrent announcement that it has formally reclassified AI investments from experimental R&D to core infrastructure — with a 2026 technology budget of approximately $19.8 billion and 2,000 dedicated AI staff — provides enterprise context: the world's largest investment bank treats AI governance as infrastructure, not a compliance afterthought.

Technical Deep-Dive

Agent 365's architecture reflects the layered security model that Microsoft has applied to endpoint and identity management for over a decade. The system operates through three distinct layers: discovery (finding agents), inventory (cataloguing what they are and what they can access), and governance (enforcing policies about what they are permitted to do).

Discovery uses a combination of endpoint signals from Intune-managed devices, cloud API integrations with AWS and Google, and network traffic analysis to identify agents operating in the environment. Local agents — programs running on employee machines like Claude Code — are detected through filesystem signatures and process monitoring rather than network traffic, which is why Intune integration is required for local agent discovery.

The asset context mapping feature, releasing in public preview in June 2026, will generate a graph for each discovered agent showing the devices it runs on, the MCP servers it is configured to use, the user identities associated with it, and the cloud resources those identities can reach. This graph-based view makes it possible to understand the full blast radius of a compromised agent credential — critical information for security teams assessing agent deployment risk.

Policy enforcement is implemented through Intune's existing device management infrastructure, meaning enterprises that already manage employee endpoints with Intune can add agent governance without deploying new agents.

The ASEAN Perspective

For IT administrators in Singapore, Malaysia, and Indonesia managing Microsoft 365 enterprise environments, Agent 365 fills a specific gap: the visibility and control layer for AI tools that employees have adopted without formal IT procurement. The shadow AI challenge is acute in ASEAN enterprises where individual employee adoption of AI tools has significantly outpaced IT governance frameworks.

Singapore's Monetary Authority (MAS) has been increasingly explicit about AI governance expectations for financial institutions. Agent 365's audit trail capabilities — logging agent actions, resource accesses, and identity associations — provide the documentation infrastructure that financial sector compliance requires.

The multi-cloud agent inventory capability is particularly relevant for ASEAN organisations that have adopted a multi-cloud strategy across AWS, Azure, and Google Cloud. Managing agent governance across all three platforms from a single console reduces the administrative overhead that would otherwise discourage deployment.

RECATOOLS Verdict

The ability to block Claude Code through an Intune policy is a headline feature that will generate significant discussion among developers — but it reflects a legitimate enterprise governance requirement, not a competitive move by Microsoft.

Enterprises have a genuine need to understand which AI agents are operating in their environment, what data those agents can access, and whether they comply with security and regulatory requirements. Agent 365 provides infrastructure for answering those questions at scale.

For ASEAN IT leaders, the practical value of Agent 365 is in the audit trail and multi-cloud inventory capabilities rather than the blocking feature. Understanding what agents are running across your cloud environments is prerequisite to governing them safely.


Frequently Asked Questions