AI Is Already Inside Microsoft 365. Most Teams Don't Know It.

Something shipped into PowerPoint yesterday that wasn't there on Monday.

At Build 2026, Microsoft announced seven new in-house AI models under the MAI brand. Most of the coverage went to the developer tools: a reasoning model, a coding model, a new transcription engine. The detail that mattered more for anyone running a business is that MAI-Image-2.5, Microsoft's new image generation model, is already live in PowerPoint and rolling out to OneDrive. You didn't have to sign up for it. If you have a Microsoft 365 subscription, it shipped.

That's worth knowing on its own. Open PowerPoint today and spend a few minutes finding every AI feature in the menus. Not to evaluate them, just to know what's there. Many teams are already using AI capabilities they never consciously switched on, because platform vendors made integration the default.

But the more consequential announcement from Build wasn't the image model. It was Microsoft Scout.

Scout is a new personal agent for work, currently rolling out to a limited group of customers, with broader availability coming. Microsoft described it in the official announcement as a system that "proactively handles things like meeting prep, scheduling conflicts and routine tasks without asking." It runs inside Teams and Outlook. It uses the context of how work actually happens across your email, your meetings, and your calendar to take action before you initiate anything.

The phrase "without asking" is worth a pause.

Most of the AI tools businesses have adopted in the last couple of years are reactive. You type something, you get something back, you decide what to do with it. That's a model most people understand. Scout is different in kind. It observes how work flows through your tools and acts on your behalf. The upside is real: prep briefs generated before a client call, scheduling conflicts caught before they become problems, routine follow-ups drafted before you thought to write them.

The part that deserves some thought before Scout reaches your team is this: what happens when it acts on something sensitive? An agent that schedules without asking can reschedule without asking. An agent that drafts from your context is acting from your relationships, your history with a contact, your implied tone with a client. When it gets it right, that's convenient. When it gets it wrong, the error carries your name.

Before Scout rolls out broadly, the useful exercise is to decide one thing as a team: what categories of action are you comfortable with an agent taking without confirmation? Meeting prep is probably fine for most organizations. External communications are probably not. Somewhere between those two is where your actual boundary sits. Writing that down now, even loosely, is easier than trying to establish norms after an agent has already been running for six months.

The broader pattern at Build matters even if you're not a Microsoft 365 shop. For years, AI adoption has been framed as a decision. Do you want to add it? The framing is shifting. The software you already pay for is building AI into its defaults. The tools your team uses daily are making capability decisions on your behalf, and the question of whether to "use AI" is becoming less relevant than whether you understand what the AI in your existing tools is doing.

Pick one app your team lives in and spend twenty minutes this week mapping every AI-powered feature it includes. You're probably not going to find anything alarming. But you might find something that changes how you think about your workflows, and you'll definitely walk away with a clearer picture of what's already running in your stack.

The conversation in most businesses about AI is still stuck on whether to start. Meanwhile, the tools they already use have started for them.