If you've ever sat in a meeting where someone said "Copilot" and three people pictured three different products, you're in good company. Microsoft seems to have noticed too. On September 25, it introduced the new Copilot, organized around three tabs: Home, Code, and Autopilot. Home brings Chat and Cowork together, with Word, Excel, and PowerPoint built right in. Code lets people who have never written a line of software describe an app, a tracker, or a dashboard and have Copilot build it. Autopilot, the agent Microsoft called Scout until this week (yes, even the newest piece already has a former name), takes on recurring work in the background and, in Microsoft's words, "keeps working while you sleep."
Microsoft cleaned up what you see on the screen and moved the complexity to the invoice.
For anyone who has had to explain Copilot to a CFO, the good news is that the menu is simpler than it has been at any point since launch. The bad news sits a few scrolls further down the same announcement. Chat stays on a fixed per-user license, while Cowork, Code, and Autopilot run on usage-based billing, metered in Copilot Credits. Microsoft cleaned up what you see on the screen and moved the complexity to the invoice.
Look at what I had to piece together just to understand how we got here: six phases in three years, a brand name stretched across more than 80 things, a chat product that went through four major name changes, and a pricing model that shifted almost every time the name did. I've spent a lot time following the collaboration technology market, and I still had to take a moment to rebuild the timeline so I could gain perspective. If you felt lost along the way, that was a reasonable response to what Microsoft put in front of you.
This is the path, with a nickname for each stop along the way.
Microsoft put one name over everything. Bing Chat launched in February, and by September Microsoft was promising "a single Copilot user experience across Bing, Edge, Microsoft 365, and Windows." Microsoft 365 Copilot became generally available on November 1 at $30 per user per month, and two weeks later Bing Chat became Copilot too. A free consumer chatbot and a premium enterprise add-on now shared one name, which was the first sign that the name would be doing a lot of heavy lifting.
What you paid for: a $30 seat.
In January, Microsoft dropped the 300-seat minimum so smaller companies could buy in, and it launched Copilot Pro for consumers at $20 a month. Wave 2 in September brought Copilot Pages and the first agents. Then in December, the Copilot app for Windows stopped working with business accounts, and the consumer app and the work app went their separate ways. One brand now had two apps, and the confusion had a foothold.
What you paid for: the same seat, sold to more companies, plus a $20-a-month consumer subscription.
The business chat became Microsoft 365 Copilot Chat, free for Microsoft 365 business users, with agents billed pay-as-you-go. The Microsoft 365 app became the Microsoft 365 Copilot app. By one fan site's count, that chat product had been through four major name changes in 23 months, which is a faster pace than most of us change job titles.
What you paid for: nothing for chat, a $30-per-user-per-month seat for the full product, and a meter for agents. Remember that meter, because it comes back.
In September, Microsoft added Anthropic's Claude models to Researcher and Copilot Studio. In October, Copilot Pro quietly folded into Microsoft 365 Premium for consumers at $19.99 a month. At Ignite in November came Agent 365 for managing agents, Work IQ for work context, and a Copilot Business tier at $21 per user for organizations under 300 people.
What you paid for: Copilot, increasingly tucked inside a bundle, at a price that depended on who you were: $19.99 a month for consumers, $21 per user per month for businesses under 300 people, and $30 per user per month for everyone larger.
March brought Copilot Cowork, built with Anthropic, along with a new E7 suite at $99 per user that includes Copilot and Agent 365. A week later, Satya Nadella's memo put consumer and commercial Copilot under a single leader for the first time, a sign that the org chart needed the same tidying as the product list. At Build in June, Microsoft introduced Scout as its first "Autopilot," an agent that gets to work without waiting to be asked.
What you paid for: a premium suite at $99 per user, with agent work starting to run on credits.
In July, Microsoft retired consumer features that few people used and reported more than 30 million paid Microsoft 365 Copilot seats. On that earnings call, Nadella said "Copilot is evolving rapidly from chat to Cowork to Autopilots." In August, the consumer and business apps merged into a single Microsoft Copilot app. Then September 25 made the pricing split official, and partners report that starting November 2, new Copilot Business purchases through resellers default to pay-as-you-go for credit-consuming services.
What you paid for: a $30-per-user-per-month seat for chat (Microsoft hasn't said whether that price holds under the new model ????), and a meter for everything that does work on your behalf.
Line up those six payment models and the confusion becomes even more evident: a seat, a bigger seat, free plus a meter, a bundle, a suite plus credits, and finally a seat plus a meter. The names changed at nearly every step, and so did the way you paid.
If you want to see how far the name itself has wandered, AI consultant Tey Bannerman has built a great site that maps the 80 or so products, features, and gadgets carrying the Copilot name. It's a visually compelling, and pretty funny, view of how the Copilot name lost its way.
This is my take on why Copilot has evolved the way it has. I map it to four pressures, and none of them will surprise anyone who has sat through a Copilot rollout.
The name stopped meaning anything specific. By April 2026, Bannerman's map had counted 81 things carrying the Copilot name, including apps, features, a keyboard key, and an entire category of laptops. When a brand covers both a key on your keyboard and the laptop it sits on, it has stopped telling buyers what they're buying. Microsoft's own people noticed. Windows Central quoted one employee describing "a delusion on our marketing side where literally everything has been renamed to have Copilot in it." I couldn't have said it better.
Paid adoption lagged the installed base. In January 2026, Microsoft reported 15 million paid seats, which Directions on Microsoft calculated as about 3.3% of roughly 450 million commercial Microsoft 365 seats. By July the number had doubled to more than 30 million. That increase is sizable, but it's still only a fraction of the installed base. It also raises the question of how much of the growth came from bundling, since the E7 suite includes Copilot, and how much came from end users choosing it.
Workers picked other tools when they had the choice. Recon Analytics surveyed more than 150,000 U.S. respondents between July 2025 and January 2026. When Copilot was the only AI tool available at work, 68% of workers used it. When ChatGPT was also on offer, that fell to 18%, and when Gemini joined the lineup, it fell to 8%. Being preinstalled gets Copilot onto the desktop, but it doesn't get Copilot picked. A user interface that automatically routes work to the right mode takes one more decision off the user's plate, and one more chance to wander off to another tool.
Agent work costs more to run. A chat answer is a single request. A Cowork task, or an Autopilot running a supplier review over several days, can call a model again and again and pull in far more context along the way. Microsoft's own pricing language draws the line: a fixed license for "quick answers, first drafts, summaries and analysis," while "agentic work requires a different type of spend." A flat seat price was never going to cover an agent that works the night shift.
The first three pressures explain why the menu got simpler. The fourth explains why the bill didn't.
If you're an enterprise buyer, take the win. Fewer names, one app, and a "UI router" that picks the right mode for you all cut the training burden that slowed early rollouts. A tool that's easier to explain is easier to adopt, and adoption is where Copilot has struggled most.
Then take a closer look at the bill, because it deserves more attention now than it ever has. Under the seat model, Copilot spend was a line item you could forecast from headcount. Under usage-based billing, spend depends on which model runs, how much context it retrieves, how complex the task is, and how many tools it calls.
But now, Autopilot adds something IT budgets haven't had to handle before: spend that accrues without anyone typing a prompt. An agent that keeps working while you sleep also keeps consuming credits while you sleep. Sweet dreams.
That brings us to FinOps for AI, which cuts both ways. Microsoft describes FinOps for AI as "a shared discipline for managing AI spend and optimizing it for business value," delivered as new capabilities across Agent 365, Insights, and Microsoft Copilot. It's unclear whether this is a feature or a new product offering. Will this be the 82nd product to co-opt the word Copilot in its naming?
Nobody installs smoke detectors in a house that can't catch fire, and nobody ships a discipline for managing AI spend unless AI spend has become hard to manage.
The new controls should be useful, as long as they don't go down the path of becoming as confusing as the product matrix. Cost management in Agent 365 now covers Code and Copilot Managed Runtime, the hosting layer for the apps people build. Microsoft has announced these features:
Spending policies: admins can set spending policies at scale through an API.
Credit requests: employee requests for more credits can flow into existing approval and automation workflows.
Model families: admins can limit which model families each group can use, which also limits what Copilot's automatic model selection can pick.
Cowork outcomes: business leaders and admins can see which Cowork tasks produce the strongest outcomes.
Employee view: employees can see their own credit usage, remaining balance, and usage history in Copilot.
I'm glad these tools exist. They're also an admission. Nobody installs smoke detectors in a house that can't catch fire, and nobody ships a discipline for managing AI spend unless AI spend has become hard to manage.
Seeing the spend is the easy part. Once the insights start flowing, the question for every IT leader is whether you have the tools and the means to act on them: to regulate usage, and to roll Copilot out to employees in a way that ties spend to value. These are the questions I'd want answered before taking Cowork, Code, or Autopilot beyond a pilot group:
Who owns the budget? IT buys the platform, and the business units will burn the credits. Decide who carries the cost before the first bill arrives and decides for you.
Are the limits in place before launch? Set organization-level and per-user monthly limits and alerts first. Partner guidance also points to an "Auto-apply new services" setting that decides whether new credit-consuming services join your spending policy automatically. Check it, because defaults have a way of becoming decisions.
Which groups get which models? Model choice is now a cost decision and an accuracy decision at the same time, so make it on purpose, group by group.
How fast does a credit request get approved? A slow approval path stalls adoption, and no approval path at all breaks the budget.
What does success look like for an agent? If you can't name the outcome a Cowork task or an Autopilot is supposed to deliver, you can't tell whether the credits were well spent.
Who watches the agents that run unattended? An always-on agent needs an owner who reviews its work and its spend on a regular schedule, the same way you'd check in on a new hire.
In my opinion, Microsoft is taking the right steps to make Copilot easier to use. Moving from a fragmented product portfolio to one "UI router" that sends each request into the right AI workflow is a good thing. It should help increase user adoption and, more important, repeat use. However, there's a new burden on you: making Copilot affordable at scale is now your job, and the organizations that answer these questions before their first Autopilot clocks in will get the benefit of the simpler menu without a surprise on the bill.
Now we get to see whether one app and a FinOps approach to AI can hold the name together. Or whether number 82 is already in the works.