Top 10 Ignite 2026 Announcements That Will Change How Admins Manage Windows, Identity, and AI

AI agents are becoming a new class of enterprise resource, and Microsoft is giving admins tools to secure them. Here are the 10 Ignite announcements that will shape your priorities, from agent identity and Intune controls to Windows Server 2028.

Ignite 2026

Ignite is always a firehose of announcements, and 2026 is no exception. But having worked with Microsoft technologies for more years than I care to admit, I know that the most heavily promoted announcements aren’t necessarily the most important for IT pros.

AI agents are becoming another class of enterprise computing resource. And if agents are going to make changes on our behalf, IT needs ways to manage and secure them.

At the same time, Microsoft hasn’t forgotten about the less glamorous realities of running Windows environments.

So, here is my top 10 list of announcements that I think Microsoft admins should have on their radar.

1. Windows Server 2028 is coming

Let’s start with something familiar, a new version of Windows Server.

Windows Server 2028 is now in preview. Server Core is getting stronger security out of the box, a new security baseline, Trusted Launch for virtual machines (VMs), automated certificate management, and support for post-quantum cryptography. Storage isn’t forgotten, including NVMe over Fabrics, native NVMe storage, and the ability to boot from ReFS.

Windows Server 2028 will get full support for Microsoft Intune management, alongside Azure Arc and Windows Admin Center.

Windows Server 2028 will be able to run AI inference close to applications and data using technologies including Windows ML, Foundry Local, vLLM and Llama.cpp. It will also support accelerated AI hardware from AMD and NVIDIA.

2. AI agents are getting their own identities

Entra Agent ID will be able to assign identities to discovered local agents and Microsoft cloud agents. Initially, that includes local agents such as GitHub Copilot CLI and Claude Code.

And that’s important because just letting an agent operate under the identity of the person who invoked it isn’t enough for serious enterprise governance.

Microsoft is building more controls around this too. Entra Global Secure Access is gaining visibility and controls for agent traffic, while an MCP Firewall will let organizations control access to MCP servers, prompts, and more.

3. Intune will help IT find unauthorized AI agents

You can only govern agents if you know they exist. So, new controls in Intune, currently in preview, will let IT discover which local AI agents are running on managed devices using Microsoft Agent 365. And administrators will be able to mark devices running unapproved agents as noncompliant. Microsoft Entra Conditional Access can then prevent those devices from accessing company data.

For approved agents, administrators will also be able to define boundaries around what they can access and do using Microsoft Execution Containers. Entra Global Secure Access can extend those controls onto the network, governing how local agents reach internet, SaaS, and private resources.

4. Agent lifecycle management: Agent 365 is becoming the admin center for your AI workforce

Once organizations start deploying dozens or hundreds of agents, who is going to manage them?

A new Action Center will identify agents that need attention, including inactive agents and agents that no longer have an owner. Adding business-use-case and task classification information will allow administrators to understand what agents actually do and identify agents performing overlapping work.

Admins will also get centralized controls over plugins and skills used by Copilot and Cowork. Admins can assign availability to specific users and groups, while expanded governance for MCP servers will provide visibility into tools and their dependencies.

Interestingly, Microsoft is making some Agent 365 usage and activity information available in the Microsoft 365 admin center without requiring an Agent 365 license.

5. Defender can protect vulnerable systems before they’re patched

Vulnerability shielding in Microsoft Defender is now in preview. The technology can apply targeted temporary controls to known vulnerable systems during an active attack.

If it works reliably, that’s potentially a useful addition for security teams.

6. On-Premises Baseline Security Mode: Exchange and SharePoint Server security

Microsoft is starting a new on-premises security initiative initially covering supported Exchange Server, SharePoint Server, and Skype for Business Server.

On-Premises Baseline Security Mode is designed to build more of Microsoft’s recommended security configuration directly into its server products, reducing the amount of manual hardening. Microsoft also wants security updates to reach connected servers faster and plans temporary mitigations for emerging threats until a permanent update becomes available.

7. Intune wants to update Windows apps IT didn’t install

Automatic updates will expand to all Windows applications, not just applications deployed by IT. The capability is currently planned for private preview.

If Intune can reliably identify and update third-party applications, that could solve a real endpoint-management problem.

Intune is also getting:

  • declarative device management
  • new Windows baseline protections
  • feature rollout controls
  • and additional operating-system reliability reporting through Windows Autopatch.

8. Copilot is about to start managing files on Windows

AI-assisted local file management is in preview. It allows users to ask Copilot to find, organize, move, copy, and rename files using natural language.

We’re also getting local PC diagnostics. Copilot will be able to examine Windows diagnostic information, help determine what is causing a problem, and suggest what to do next.

Plus, easier Copilot access is coming directly into the Windows Search experience on the taskbar, with controls for IT administrators.

9. Edge can turn recorded browser workflows into Copilot skills

Lots of businesses still depend on internal web applications with poor integration options or no useful API at all. So, Microsoft Edge for Business is getting a skills recorder. Employees will be able to record a workflow in Edge and turn it into a reusable skill for Copilot Cowork. IT controls whether the capability is enabled, and recorded skills will operate within the user’s existing permissions and organizational policies.

Managed agent browsing is also coming into preview. AI agents get dedicated Edge windows and their own Microsoft Entra agent identity, with visual indicators showing when a window belongs to an agent.

10. Defender is extending protection to ChatGPT, Claude, and local AI agents

Finally, Microsoft Defender now provides generally available runtime protection for a range of local AI tools on Windows for Endpoint P2 customers. IT is no longer just securing what the user does on the PC. But also securing what software acting on behalf of that user is allowed to do.

Agent-native protection covers products including:

  • Claude Code
  • GitHub Copilot CLI
  • GitHub Copilot app
  • Claude Desktop
  • VS Code extensions for GitHub Copilot and Claude Code
  • Network-based protection also covers tools including Ollama Desktop and ChatGPT Desktop.

Defender can inspect local agent calls for these threats, with administrators choosing Audit or Block modes through Defender or Intune. The runtime protection is available to Defender for.

The bigger Ignite story is control, not AI

There’s obviously plenty more at Ignite 2026. Microsoft Perception is now generally available with specialized security agents. Windows 365 for Agents is getting persistent Cloud PCs. Microsoft IQ is designed to give agents better access to organizational knowledge and context. And Microsoft continues to expand agentic capabilities throughout Azure.

But I think concentrating only on the AI features misses the bigger story.

Microsoft is starting to build the infrastructure required to answer the much harder questions: Who is this agent? What can it access? Which device is it running on? Etc.

I think that’s more important than another Copilot demo.