Your employees can now build their own AI agents

If you’ve used Copilot in Word or Outlook over the past year, you’ve got a mental picture of what it does. It drafts an email. It summarizes a document. It cleans up a spreadsheet when you ask it to. Useful, but it waits for you to ask.

That picture just got out of date.

Microsoft announced Copilot Wave 3 in March 2026, and this one is a different kind of update than the last two. Wave 1 put Copilot inside your documents. Wave 2 spread it across the rest of Microsoft 365. Wave 3 lets your employees build their own AI agents: task-specific assistants that carry out work inside Word, Excel, PowerPoint, and Outlook without someone walking them through every step.

That’s a real capability. It’s also a decision that lands on your desk, not just IT’s. The agent your marketing coordinator builds to draft client follow-up emails is an employee who never went through onboarding, never signed anything, and has whatever access she decided to give it.

What Copilot Wave 3 actually adds

The headline feature is the ability to build an agent right inside the apps you already use. Think of something narrower than a full AI assistant: an agent that formats incoming invoices the same way every time, or drafts a first pass on routine client emails. You describe the job once, and it doesn’t need you standing over it for every run after that.

Copilot’s chat window also got more capable. You can now tell it to change a document, spreadsheet, presentation, or email directly from the conversation, instead of clicking between screens to do it yourself.

Microsoft is also introducing something called Work IQ. It’s Copilot reading your organization’s own data, past communications, and internal files to answer in a way that fits how your business actually runs, instead of a generic answer that could apply to anybody. Copilot itself is getting more flexible under the hood too. It can now run on models from Anthropic and OpenAI, not just Microsoft’s own.

One more thing worth watching, even though it’s still in preview. Microsoft is testing something called Copilot Cowork, which takes the agent idea further. Instead of responding to one request, it can run a longer, multi-step process on its own and check back in once it’s done.

Who’s watching the agents your team builds?

Here’s the part that matters more than any feature on that list. If Copilot can now build agents that take action inside your business’s own tools, without someone approving each step, that’s a security and defense question: who’s building them, what they can reach, and what they’re doing all day.

Microsoft clearly thought about this too. Alongside the agent-building tools, it’s launching Agent 365, a management platform built specifically to let a business see and govern the AI agents its employees create. That’s Microsoft telling you, indirectly, that ungoverned agents are a real problem, not a hypothetical one.

Think of it this way. An AI agent with access to your files and your inbox is functionally a new employee. Would you let a new hire loose on client email and financial records with no manager reviewing the work and no idea what they’re doing when you’re not looking? Probably not. An agent built on a whim, with no oversight, amounts to exactly that.

What to do before you turn this on

Agent 365 is set to launch May 1, 2026, priced at $15 per user. Microsoft is also bundling it into a new top-tier package called the Frontier Suite, at $99 per user, alongside Copilot, Entra, Defender, Intune, and Purview. That bundle is built for larger, security-heavy organizations, and the price reflects it. For most businesses, the practical move isn’t buying the biggest package Microsoft offers. It’s deciding, before anyone on your team starts building agents, who’s allowed to build one, what data it can touch, and who reviews what it’s doing.

If your business is already on Copilot and your team wants to try building agents, that’s fine. Set the rules first. Decide which roles can create them. Decide what systems they can reach: client records, financial data, email. Decide who checks in on them, whether that’s Agent 365 or a simpler review process your IT provider sets up.

If you’re not on Copilot yet, this isn’t a reason to rush in either. It’s a reason to have someone who understands both the tool and your business look at what makes sense before you turn any of it on.

Copilot went from an assistant that waited for you to ask, to a tool that can hire its own help. That’s a bigger shift than it sounds, and it’s worth making on purpose instead of by accident. If you want help sorting out which of these features are worth turning on, and setting up the right guardrails before your team starts building agents, get in touch.