Search

With The New Copilot, Microsoft Starts Selling Work, Not Just Seats

Satya Nadella

Enterprise AI is shifting from chat to delegation, and the major AI platforms are racing to own that handoff. Anthropic has Claude Cowork, OpenAI has ChatGPT Work, and Amazon has Quick. Microsoft answered Friday with a rebuilt Copilot. CEO Satya Nadella and Jacob Andreou, the EVP who runs Copilot, showed the app to customers in Seattle on Wednesday, and Jared Spataro, Microsoft’s CMO for AI at Work, detailed the launch in a blog post. The app adds Home, Code, and Autopilot and builds in Word, Excel, and PowerPoint. I use these agentic tools regularly and have complained publicly when vendors make simple things hard, so I looked for what Microsoft can do that the model labs can’t.

The answer is the business model, not the tabs. In March, I said Microsoft 365 E7 could be the most significant commercial packaging change since E5. This change is bigger because it rewires how Microsoft gets paid for AI: a seat for everyday work, a meter for agent work. My non-consensus take is that this is margin repair and market expansion, not a price cut, and it’s the right call. (Disclosure: Microsoft is a client of Moor Insights & Strategy, my firm.)

What Microsoft Announced With The New Copilot

Home puts Chat and Cowork in one place. Code is designed to build apps, dashboards, and workflows from plain-language descriptions, using GitHub Copilot technology, and to host them in the tenant on Copilot Managed Runtime. Autopilot, formerly Scout, is a persistent agent that gets its own identity, memory, computer, and workspace in the tenant. Microsoft says Home and Code reach Frontier early-access customers in the coming weeks and Autopilot enters private preview this month.

Why Usage-Based Billing Is The Real Copilot News

The per-user license covers Chat, Copilot in the Office apps, model selection, and Auto, which picks each request’s model by accuracy, speed, and cost. Cowork, Code, Autopilot, and frontier models such as Astra and Fable run on usage-based billing in Copilot Credits, which requires the seat.

That’s the GitHub Copilot playbook applied to Microsoft 365. GitHub Copilot moved to usage-based pricing in June. On the July earnings call, Microsoft said GitHub Copilot revenue accelerated more than 60% quarter over quarter, and CFO Amy Hood said margins improved after the change. Nadella told investors the business model is moving “beyond per-seat to per-seat-plus-consumption.” The Information reported this week that Microsoft is authorizing 30% to 50% seat discounts for large buyers committing to usage fees. Lower seat, bigger meter.

My thinking has moved. In January 2025, I called pay-as-you-go a step in the right direction to prove ROI and then flip to per seat. Microsoft did the opposite: E7 made the seat the default, and the meter sits on top. For agents, that order is right: agents don’t have headcount. That order also answers my colleague Jason Andersen’s April worry about the lack of a consistent anchor in activity-based pricing: the seat is the anchor.

Why Governance Is Microsoft’s Copilot Moat

I also argued in March that “the real play is governance, not generation.” Now governance is also the billing system. Usage-billed services stay off until an admin creates a spending policy. Policies set tenant, group, and user budgets, limit model families by group, and route credit requests into approval flows.

The model is now a component. Both tiers run on OpenAI and Anthropic models today, with more labs and open-weight models coming. At Wednesday’s event, Nadella said true independence means having your own private tests for models and the ability to “substitute a model.” That’s the architectural disconnection between the model and every other stack layer I flagged in July. Microsoft’s edge is the identity, security, and business context already in Entra, Teams, Outlook, and SharePoint.

What Microsoft Must Still Prove On Copilot Costs

The number that matters is cost per correct outcome, and buyers don’t have it yet for Copilot’s agents. Credits cost a penny each, and Microsoft published task patterns when Cowork shipped in June. In the Seattle demo, an Autopilot named Dot watched a retail manager’s Teams, email, and inventory data and flagged a shipment problem. Microsoft hasn’t said what an always-on agent such as Dot, or a Code app, burns in credits. Andreou told CNBC the product optimizes for output quality over cost efficiency. Good for users; cost discipline falls on the customer.

Accuracy costs money. One of my companies, Signal65, ran Claude Fable 5.1 through the PINNACLE benchmark on September 2, in thinking mode over the Anthropic API. Fable invented answers to 0.7% of questions the documents couldn’t answer, versus 7.6% for Claude Opus 5 and 10.3% for GPT-5.6 Sol. It also cost $2.46 per correct task, 1.8 times Opus 5. As of September 25, OpenAI’s GPT-6 Astra at maximum effort tops PINNACLE’s live results. Microsoft meters Astra and Fable and puts GPT-5.6 in the seat, so the best-scoring models in Copilot are the metered ones.

Microsoft’s own cost comparison against Claude and ChatGPT uses modeled list prices; treat the comparison as a vendor claim. Spending policies also auto-apply to new metered services by default; turn that off to review services before they bill. Andersen called layered seat-plus-consumption billing pricing that resists planning.

What Microsoft Should Show At Ignite

Microsoft reported over 30 million paid Copilot seats in July, about 7% of the 450 million-plus base I cited in March, up from 3.3%. Consumers are next: Spataro said consumer Microsoft 365 plans will add usage-based billing later this year for Cowork, Code, and more. At Ignite in November, Microsoft should publish credit burn for Autopilot and Code, make new metered services opt-in, and show customer ROI by task.

Microsoft’s Copilot Model Is Right, But The Bill Needs Proof

Office sold seats. Copilot is built to sell work: the seat gets customers in, and the meter is meant to grow the business. Microsoft has the right business model and the right governance. It hasn’t yet shown that customers can predict the bill or that the agents get the work right.

Sources

Patrick Moorhead
Founder, CEO and Chief Analyst |  + posts

Patrick Moorhead is the founder, CEO, and chief analyst of Moor Insights & Strategy. His big-picture view of technology is grounded in more than 20 years as an executive leading strategy, product management, product marketing, and corporate marketing functions at NCR, AT&T, Compaq, and AMD. He has shared his expertise in areas from silicon to infrastructure to enterprise SaaS and everything in-between in thousands of national broadcast appearances (CNBC, Yahoo Finance), articles (Forbes, CIO), research-based analyses, and podcast episodes. Today, he has 100+ CXO-level advisory clients and is often ranked the #1 technology industry analyst by ARInsights.

Recent Posts

IMG_5450

Microsoft Fabric Puts Power BI’s Business Context to Work at FabCon Europe 2026