AI Agents Are Getting Employee Badges: What Microsoft's Agent 365 and Autopilot Mean for Your Business
Microsoft held its Ignite conference in San Francisco this week, and the message from the keynote was hard to miss. Satya Nadella called Copilot "the new operating system for work", and almost everything announced after that treated AI agents not as chat windows but as something much closer to staff. Agents are getting their own identities, their own managers, their own budgets and even their own risk ratings.
Three names matter most. Agent 365 is a control plane for managing whole fleets of agents across a company. Autopilot is a personal agent that works continuously in the background without being prompted. And Microsoft IQ is the layer that feeds agents your company's own knowledge and data. Microsoft says more than 50 million agents are already managed through Agent 365, which gives you a sense of how fast this is moving.
Even if your company does not run on Microsoft, these announcements set a pattern that the rest of the industry tends to copy within a year. Here is what actually launched, what it costs, and what you should do about it. Let's sink in!

What Microsoft announced at Ignite 2026
Ignite runs from 17 to 20 November, in San Francisco and online, and the first day was packed. Copilot got a new Home, a single entry point that combines chat and collaborative work, with an automatic router that picks the model and effort level for each request. Companies can let employees choose between OpenAI and Anthropic models within whatever policy IT sets. Office in Copilot, which creates Word, Excel and PowerPoint files from plain descriptions, moved to general availability and is included in the Copilot licence without usage metering.
Around that core, Microsoft introduced Copilot Workspaces, a shared space where teams and agents work with the same context, and Today, a daily brief of priorities and prepared actions. Then came the bigger structural pieces: Autopilot, Microsoft IQ, an expanded Agent 365, and a security platform called Microsoft Perception that is now generally available.
Nadella framed all of it with three principles: keep humans at the centre of work, let organisations own their intelligence, and build trust through governance. The third one is where most of the practical news sits.
Agent 365: a staff directory for your digital workers
Agent 365 is best understood as an HR and IT system for agents. It extends the identity, security and management tools companies already use for people to the agents they deploy, whether those agents come from Microsoft, from third parties or from internal projects. Every agent gets an identity in Microsoft Entra, an owner, and a place in the inventory.
The part finance teams will like is Cost Management, which is available now. It tracks spending on usage-based AI services by user, by agent and by model, and lets admins set limits and alerts. A related Business Analytics feature, still in the early-access Frontier programme, tries to tie AI spend to work completed and estimated human time saved, which is the ROI question every board has been asking.
The governance side is just as interesting. There is a risk engine that rates each agent low, medium or high risk, detection of shadow agents that nobody registered, and an action centre for inactive or ownerless agents coming in preview in December. Full governance and security requires an Agent 365 licence or the new Microsoft 365 E7 tier, so this is also where the money is.
Autopilot: an agent with its own ID and its own manager
Autopilot is the announcement that generated the most discussion, and for a good reason. It is a personal agent that runs continuously in Teams, Outlook and documents, picking up recurring work and moving projects forward without being prompted each time. It started rolling out to Frontier programme customers on 17 November, with no date yet for general availability.
What makes it different from earlier assistants is the accountability model. Each Autopilot instance has its own Entra Agent ID, its own memory and tools, and a named human manager who is responsible for it. IT can scope what it reaches or block it entirely. In other words, the agent appears on the org chart, reports to someone, and can be audited like an employee.
That design answers the question that has stalled many agent projects in real companies: who is responsible when the agent does something wrong. Microsoft's bet is that agents become adoptable at scale only when someone's name is attached to each one.

Microsoft IQ: feeding agents your company's context
The quiet workhorse of the announcements is Microsoft IQ, which is generally available. It gathers organisational context from four sources: Work IQ covers email, meetings, documents and chats; Fabric IQ covers business data and metrics; Foundry IQ covers unstructured documents; and Web IQ brings in public web information. There are more than 350 third-party connectors, and no standalone subscription, since it builds on Microsoft 365, Fabric, Foundry and Azure.
The strategic argument behind it is worth taking seriously. Microsoft is saying that competitive advantage will not come from access to powerful models, because everyone has that. It will come from how well a company captures and governs its own knowledge. An agent with a brilliant model but messy, outdated internal documents will still produce messy, outdated work.
For most businesses this is the least glamorous and most useful takeaway of the whole event: the quality of your SharePoint folders, your permissions and your documentation is about to become the quality of your AI.
What it costs, and what is actually available
Microsoft published no list prices at Ignite, so budget figures will have to wait. What is known is the structure. Office in Copilot and the Today brief are included in the Microsoft 365 Copilot licence. Autopilot, Copilot Code, Cowork and the Today feed consume Copilot Credits, which are usage-based. Full Agent 365 governance sits behind an Agent 365 licence or Microsoft 365 E7.
Availability comes in three buckets. Generally available today: Office in Copilot, Microsoft IQ, Perception, the core of Agent 365 Cost Management, and the Copilot Managed Runtime. In the Frontier early-access programme: Autopilot, Copilot Code, Workspaces and Business Analytics. In preview: Today, the Salesforce plugin and several Entra and Purview controls.
The credits model deserves attention before anything else. Usage-based pricing means variable bills, and variable bills plus autonomous agents is a combination that can surprise you. Setting spending limits and alerts should come before enabling any credit-consuming feature, not after the first invoice.

What this means for businesses building digital products
Microsoft was not alone this week. Replit launched Missions, which coordinates several agents toward a defined outcome with spending limits and approval rules, and Cognition introduced Devin Bots, persistent agents that hold context on a software project for weeks. Three companies shipping the same idea in the same week is not a coincidence; the industry is converging on managed fleets of long-running agents.
For a business, the practical to-do list is fairly clear. Make an inventory of the agents and automations already running in your company, including the unofficial ones someone set up with an API key. Give every one of them a named human owner. Put spending limits in place before rolling out anything usage-based. And invest in documentation and data hygiene, because context layers like Microsoft IQ can only work with what you give them.
If you build web products, there is one more step: expect agents among your users. An agent booking a service or comparing suppliers reads structure, not design. Clean markup, working APIs and accurate product data are becoming a sales channel of their own, and testing how an agent experiences your site is worth doing now, the same way you once tested a new browser.
Final notes
Ignite 2026 marks the point where AI agents stopped being a feature and became a workforce category, with identities, managers, budgets and risk scores. The governance tooling, not the agents themselves, is the real news, because it is what makes large-scale adoption plausible for ordinary companies.
There is no need to chase every Frontier preview. Start with what is generally available, pilot with a small group, and spend the saved energy on the unglamorous work of owners, limits and documentation. The companies that do that quietly this year will be the ones for whom agents actually work next year. If your systems or website need preparing for that, it is better to start before the agents arrive than after.





Comments