I read the Ignite 2026 Book of News so you don’t have to, at least the Intune, Autopatch and Windows 365 parts. The short version: it’s an AI agents story, and Intune’s role in it is governing the agents that now run on your endpoints. Around that you get a pile of Windows management work, a patching push that hides in Microsoft Perception, and a long Windows 365 list.
Two things before we start. First, no Intune Suite, no Endpoint Privilege Management and no Configuration Manager in the text I read, so if you came for those, sorry. Second, almost everything here is preview or private preview. I flag the GA items because there are so few of them. Private preview means most of us can’t touch it yet, so I read those as direction, not as something to put in a rollout plan.
Intune as the bouncer for local AI agents
This is the headline Intune item, and it’s in preview. Intune discovers agents running on managed devices through Microsoft Agent 365. A device with an unapproved agent gets marked noncompliant, and Entra Conditional Access then keeps it away from company data. Agents you do approve live inside Intune policies and Microsoft Execution Containers, with Global Secure Access extending the control to the network.
My take: it’s the compliance-to-Conditional-Access chain we already use for OS versions and encryption, with “which agents are installed” as the new input. I like that. It’s a model I understand and can explain in one sentence. What the book doesn’t say is how discovery actually works or what counts as an agent, and that’s the first thing I’d test the day it lands.
On the Windows side, process container controls are in preview and learning and permissive mode is GA. That second one deserves more attention than it gets. It lets you watch what an agent does before you enforce anything. Audit first, block later. Nobody wants to learn on day one that some developer tool quietly depended on the thing you just cut off.
Then there’s Defender. Runtime protection for local AI agents can be configured in the Defender or the Intune portal, in Audit or Block mode, and it’s GA on Windows for Defender for Endpoint P2 customers. Agent-native protection covers Claude Code, GitHub Copilot CLI, the GitHub Copilot app, Claude Desktop and the VS Code Copilot and Claude Code extensions. Network-based protection covers OpenClaw, Ollama Desktop and ChatGPT Classic and Desktop. Is this the same machinery as the Intune discovery feature? The book doesn’t say, so I won’t pretend to know.
Windows management: describe the outcome, let Windows hold it
This section reads like a list of small bets that all point the same way: configure less, state what you want more.
Declarative device management (preview). IT defines the desired outcome, and Windows applies and maintains it. If you’ve worked with Apple’s declarative management, this will sound familiar, though the book never makes that comparison. It also doesn’t say which settings are covered, and that’s the first thing I’d want to know.
Windows baseline protection (private preview). Critical security settings are applied and kept in place from boot, so there’s no gap between first start and policy arriving. I’d like that gap closed.
Experience controls (private preview). IT can evaluate and roll out upcoming Windows features on its own schedule instead of whenever Windows decides. Anyone who has watched a new feature land on a Monday morning will see the appeal.
Expanded automatic updates in Enterprise App Management (private preview). Automatic updates stretch to all Windows applications, not only the ones IT deployed. That includes whatever users installed themselves, which is where unpatched software tends to hide.
Windows Endpoint Security Platform (GA). A safer way for approved security products to protect devices. The description is thin, and I read it as aimed more at security vendors than at admins, but that’s my interpretation.
Patching, with agents in the loop
Windows Autopatch gets OS reliability reporting (private preview): reliability insights, stop error trends and the affected devices. That’s the data you want the moment someone asks whether this month’s update is actually healthy.
Two agent-driven patching items are in the book as well. The patching and vulnerability management initiative (preview from January) has agents assess exposure to the monthly Windows Update vulnerabilities and coordinate permanent remediation through Intune. The Risk Reduction Agent in Microsoft Perception (preview) does something similar, also through Intune. Both stress that endpoint admins keep control of deployment, which is the sentence I’d read twice. Both come from Microsoft Perception, which gets its own section below, and that’s the part I wouldn’t skip.
Servers show up too. Windows Server 2028 is supported in Intune and Autopatch for client-like scenarios, alongside PCs (private preview). Separately, it can be managed more consistently across datacenters, cloud and remote sites with Intune, Azure Arc, Windows Admin Center and vMode, in preview through the Windows Server Insider Program.
Microsoft Perception: the part nobody tells Intune admins to read
If you only read one section outside the Intune parts, make it this one. Microsoft Perception (previously Project Perception) is now GA, including the agents and playbooks announced in July. It’s a security product, not an Intune one, which is exactly why Intune admins can walk straight past it. But it’s where the agent-driven patching comes from, and it ends in your Intune tenant.
So what is it? A security system built from specialized agents and multi-agent workflows for common security tasks, with playbooks coordinating the agents on broader challenges. The agents have roles: red team agents look for weaknesses from an attacker’s perspective, blue team agents triage alerts and investigate attacks, and green team agents help prioritize weaknesses.
The Intune connection is the Reduce exposure risk playbook (preview). It calls the Risk Reduction Agent to assess ways of reducing risk and recommend a response. When a permanent fix is ready, the agent can coordinate patching through Intune, and IT keeps control of deployment. When patching isn’t possible right away or could disrupt the business, the agent can apply targeted shielding until the fix is in place. That shielding bit has no Intune-side description in the book. Who approves it, and where do I see it? Those are my first two questions.
The patching and vulnerability management initiative from the previous section sits under Perception’s AI Readiness Initiatives (preview in January), next to modernizing authentication with phishing-resistant passkeys for administrators. Whether the initiative and the playbook are one feature or two isn’t stated.
Two practical notes. Playbooks can be triggered automatically or launched from Microsoft Defender, and APIs plus Model Context Protocol support let you start them from ITSM, SIEM and SOAR tools. And the book mentions least-privilege access, approval controls, audit logging, usage views and a cost calculator, but no licensing terms or pricing. I’d find that out before promising anyone anything.
Windows 365 gets the longest list
The admin-facing items first.
Windows 365 admin assistant (preview). An AI experience in Intune to monitor Cloud PC health and speed up troubleshooting.
Windows MAM (preview). Security and data protection policies apply to the Windows App, while the devices used to reach Cloud PCs can stay unmanaged. For BYOD and contractor scenarios this is the one I’d look at first.
Context-based redirections (GA). Clipboard, drive, printer and USB redirection are allowed or restricted based on how trusted the connection is, using Entra authentication context.
Direct Stream (GA). More resilient connectivity for Cloud PC users.
Then the AI and agent side, where it gets a bit much.
Local AI on Cloud PCs (GA). Cloud PCs with up to 32 vCPUs and GPUs can run local models through Foundry Local, with predictable subscription costs. GPU improvements built with NVIDIA are in preview early next year in select regions.
Windows 365 for Agents. Now eligible for Copilot Credits, the usage-based billing currency, so AI spend lands in one place. In preview: the Copilot Studio computer-use tool (Microsoft Copilot harness in December), a Foundry MCP option (early next year), access to on-premises apps through agent identities, and persistent Cloud PCs in Frontier.
Windows Server 2025 with AVD Hybrid (private preview). Multi-user AI development environments managed through Windows Server and Azure Virtual Desktop.
Two more worth a line
Copilot on Windows gets local file tasks and PC diagnostics (preview). The part that’s yours: admins can manage the new Windows Search experience, including Copilot access, through policy.
And Microsoft Edge for Business adds Entra Global Secure Access (preview). Employees on Windows PCs reach private web apps through their signed-in work profile, with no separate VPN client, network agent or browser extension.
What’s missing
Security Copilot for Intune, Endpoint Privilege Management, Mac, iOS and Android management, the Intune Suite, Configuration Manager and Entra device identity as its own item: none of them turn up in the parts of the Book of News I read. That doesn’t mean nothing was announced. The book is a summary, and the Intune team’s own blog posts during Ignite are the next place I’ll look. Licensing is almost never stated either, so check before you promise anything.


