Microsoft stopped updating its Copilot seat count. It would rather talk about 50 million agents.

Microsoft's Copilot seat count went quiet this week, and it took the last question of the earnings call to establish that. In July the company told investors it had "over 30 million paid Microsoft 365 Copilot seats," with net seat additions more than doubling from the prior quarter. On October 28, Tal Liani of Bank of America asked, "Did you disclose the Copilot numbers this quarter?" CFO Amy Hood answered at the end of her reply: "we did not disclose, to answer your direct question, an update on the Copilot number this quarter. We talked about momentum in revenue and usage."

The transcript gives other numbers. Hood said E7, the premium bundle, "has already reached millions of paid seats," and that Copilot revenue growth accelerated from the prior quarter. Satya Nadella said: "Over 50 million agents are now in the registry, up over 30 percent quarter over quarter." The earnings release reports "Agents and Infra" revenue of $76.8 billion, up 25 percent.

Why the Microsoft agent registry is not an adoption number

The Microsoft agent registry is not an adoption number, even though it sounds like one. Nadella introduced it while describing Agent 365 as a platform to manage, secure and optimize agents. A registry records that an agent exists. It does not say whether the agent ran last week, whether a person uses it, or whether it does paid work. A single test environment can register many agents, and Microsoft did not say how many are active.

That does not make the number useless. Growth of 30 percent in a quarter suggests companies are creating agents quickly and bringing them under management. It tells you about governance demand. It says little about value delivered.

Three companies, three ways to count enterprise agents

Three companies now count enterprise agents in three different ways. Microsoft moved from seats to a registry. ServiceNow's October 28 release raised its 2026 AI contract value target "from $1 billion at the start of the year to $1.7 billion now" and said agentic "year-to-date consumption in production" is up "more than 13x year-over-year," without giving the base. Salesforce reported in August that Agentforce ARR exceeded $1.5 billion, on a definition that now also includes other AI offerings, Slackbot and Headless 360, so it does not line up with earlier quarters.

Each number is accurate on its own terms. Together they show an industry that has not agreed on what adoption means. Vendors are also picking the unit that flatters the current moment, which is what public companies usually do. An analyst had to prompt Microsoft to say a metric had been dropped.

What the Microsoft call does not establish

What the Microsoft call does not establish is that seat growth has slowed. Hood did not say so, and Copilot revenue growth accelerated by her account. A company can stop reporting a number for many reasons, including a change in how it sells. She pointed to balancing "traditional SKU per-seat rhythms" with "the newer premium E7, M365 Copilot, and usage-based productivity offerings," and GitHub's move the same day shows usage taking over from seats elsewhere in the company. We do not know whether the seat count will return.

How enterprise buyers should read agent adoption numbers

Enterprise buyers should read agent adoption numbers the way they read any vendor metric, and ask for the denominator. Four questions cut through most of it:

• How many agents ran in production in the last 30 days?
• How many tasks did they finish without a person stepping in?
• What did each finished task cost, including model and platform fees?
• How does that compare with customers of our size?

The self-reported numbers that stand up best name a task and a volume, as Paychex did with its payroll voice agent. Those are worth more than any registry total.