November 19, 2025
By Bob O'Donnell
While most companies are still struggling with how to best integrate AI into their organizations, it’s clear that the businesses building the tools for these companies have quickly moved on to agents as the primary means of leveraging the capabilities of artificial intelligence. This was abundantly clear at the keynote for Microsoft’s Ignite event in San Francisco, where the Redmond-based giant unveiled numerous ways to create, integrate, and manage AI-powered agents into its Microsoft 365 suite of productivity apps as well as Windows 11.
Most of the reason for the focus on agents is straightforward—it’s about turning intelligence into action. By leveraging the reasoning capabilities of today’s most advanced AI models in conjunction with the context of a user’s files, emails, calendar events, chats, etc., the thought is that you can start performing useful work on behalf of an individual. It’s an intoxicating idea that provides a great example of the potential power of applied AI—as long as the reality can live up to the hype (which, in some early efforts, hasn’t always been the case).
A critical part of making this vision work is having a solid foundation of context onto which potential applications can be applied. At Ignite, Microsoft introduced a new contextual layer of intelligence it’s calling Work IQ that provides that context. While not immediately clear, it turns out Work IQ builds on top of the data in the existing Microsoft Graph. In other words, as a major LLM succinctly explained, “Work IQ essentially takes the ‘what’ from Graph and transforms it into the ‘why’ and ‘how’ for organizations.”
With the Work IQ layer in place, the options for integrating agents into the experience are manifold. To start, it’s possible to leverage prebuilt agents either in the Copilot chat experience or directly in Word, Excel, and now PowerPoint via new Agent modes. So, depending on which way you prefer to work, you can get access to tools that can help you create Office documents with simple prompts and then the necessary follow-up questions to get files with the content and format you want. It’s great to see Microsoft offer this degree of flexibility, but it can be a bit confusing for people who are still trying to figure out how to integrate these AI-powered tools into their everyday workflows.
If you want to create custom agents, Microsoft offers several ways to do that, many of which were updated at Build. Copilot Studio continues to be the tool of choice for power users and “citizen developers” to create their own agents, but the company also discussed Microsoft Foundry (the new name for Azure Foundry) for professional developers, as well as a new Agent 365 SDK (Software Development Kit). Important additions for these latter tools include the ability to run these agents in their own separate environments—not just containers—giving IT managers more opportunity to manage and control them—a point to be discussed more a bit later.
Agents built with Microsoft Foundry can now (at least in preview) also maintain a memory across sessions. This provides critical historical context as people use the tools for multiple tasks. Plus, it’s a capability that most people have grown accustomed to with the big LLM chatbots. Microsoft also made a big announcement with Anthropic (and Nvidia) about the companies working together. One of the direct outcomes of this collaboration is that Microsoft now includes support for Anthropic’s Claude, Haiku, and other models as an option when building agents. This provides yet more flexibility for developers to try out different underlying AI models as they build out their agents.
Building on the concept of Work IQ, Microsoft also announced Foundry IQ, which essentially extends the conceptual connections that Work IQ brings to individuals across the entire business. This allows agents built with Foundry to leverage all the available enterprise documents and resources in places like Sharepoint directories via a RAG-style implementation. This, in turn, makes the agents more grounded in company data (and, therefore, less likely to hallucinate).
In addition to the traditional Office apps of Microsoft 365, the company also introduced several new agentic capabilities within Windows 11. Windows has had the ability to do a semantic search and to change certain Windows settings via the Copilot app for a few months now. At Build, the company announced a new core set of services and capabilities within Windows 11—including an integrated MCP server and what the company calls an Agent Framework—that turn it into what the company initially dubbed an “agentic OS.” While that description isn’t likely to stick, it does highlight how the company is working to make Windows 11 a more friendly environment in which to create and run agents.
Leveraging Work IQ, the company showed how the latest version of Copilot can look across emails, calendar invites, chats, et al as it responds to questions users may have about finding information, planning for a meeting, etc. Not surprisingly, Microsoft was quick to point out that IT shops and users can choose which set of agentic capabilities they want to run (or block) in Windows, but the foundation for agents in Windows is now being put into place.
Microsoft also brought a new means to run certain types of agents in virtual desktop environments via Windows 365. Windows agents can leverage the company’s cloud-based Windows 365 virtual machines to create separate environments in which agents perform their tasks. This Is conceptually similar to what Foundry brings to agents that run across an entire enterprise and can help with both manageability and security of those agents. Another potentially very useful capability coming soon to Windows is the ability to use the Edge browser with an agent to do things like fill out online forms and other computer usage (CUA—Computer Usage Agent) applications.
Of course, given all these tools for building and deploying agents, it’s not hard to imagine scenarios in which agents could potentially run amok. To help avoid this, the company debuted a new tool called Microsoft Agent 365, which it described as a “control plane for agents.” While it may sound like a relatively straightforward extension to the company’s other agent-focused offerings, it’s interesting on several levels. First, it essentially treats agents as “digital employees” complete with different sets of access rights, digital IDs, and more. Because of that approach, Microsoft integrated its Defender, Entra, and Purview tools as part of Agent 365. Until now, these tools have only been applied to people, but applying them to agents allows IT managers to see and manage the activities of agents just as they do regular employees. This, in turn, provides a set of governance and security controls that can help with potentially nefarious digital agents exfiltrating proprietary corporate data or performing other harmful tasks. While the anthropomorphizing of digital agents does feel a bit odd, given all the capabilities they will soon have, it only makes sense.
Microsoft also debuted a new Foundry Control Plane that’s designed for developers who are creating agents with the new Foundry tools. While the tools might initially seem to overlap, the best way to understand the distinction between them is to know that Agent 365 is how the enterprise sees and governs agents, while Foundry Control Plane is how developers see and govern agents. It’s a lot to put together, but at the same time, it does reflect how thoroughly Microsoft has thought about the requirements for building and deploying agents across a wide range of users and roles throughout the enterprise. Of course, it also shows how complicated all of this still remains.
We’ve seen many different companies discuss their agentic efforts recently to help position themselves as forward thinkers, and in some ways, Microsoft essentially did the same thing at Ignite. However, once you dig in, you can see that the range and depth of the agentic-based offerings that Microsoft discussed at the event go well beyond what virtually any other company has yet to unveil. In all honesty, that can make the overall message difficult to decipher, and I think the company still has a lot of work to do to simplify and clarify how all its tools fit together. But there’s no question that Microsoft is placing a big stake in the ground when it comes to extremely wide deployment of agents in business environments. Now, let’s just hope that businesses can keep up with the pace of these announcements.
Here's a link to the original column: https://www.linkedin.com/pulse/microsoft-brings-agents-office-windows-bob-o-donnell-4aztc
Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on LinkedIn at Bob O’Donnell or on Twitter @bobodtech.
| 
