Workday's 5,500 agent customers, and what the count leaves out

The number counts customers, and that is all it counts

Workday's fiscal 2027 second-quarter release, published on August 27, 2026, says more than 5,500 customers now use at least one of Workday's organic agents, up 35 percent from the prior quarter, and that AI drove more than 25 percent of new annual contract value. Both figures come from Aneel Bhusri's quote in the release. Both will be on a slide at your next account review, so we want to be precise about what they measure.

We read the 5,500 the way we'd read any adoption figure in an earnings release. It counts customers, and the threshold is one agent. A tenant that switched on a single agent in one business unit counts the same as a tenant running several across finance and HR. The count folds those two together, and nobody outside Workday can pull them apart.

What the count can tell an implementation team

It tells you the enablement path works at scale. More than 5,500 customers have been through whatever provisioning, security review, and configuration steps stand between a Workday customer and a running agent, and Workday support has fielded the tickets that followed. If your implementation partner says they have never seen one go live, that is now a partner problem rather than a product one.

It also tells you the release cadence will keep favouring agents. A quarter-on-quarter rise of 35 percent gets reported on the next call, so whoever plans the Workday roadmap has every reason to ship more of them and to make the first one easier to switch on. Our sibling piece on the product changes sitting under the revenue line lists what has already shipped, including a Financial Audit Agent the release says is now generally available.

The word organic is doing quiet work in that sentence. The release doesn't define it. Our reading is that it means agents Workday builds and ships, as distinct from ones a customer or partner assembles. If that's right, the 5,500 says nothing about customer-built agents, which is the category most implementation teams are being asked to plan for this year.

What the count cannot tell you

It cannot tell you which agent. One customer's single agent might be the Financial Audit Agent reading controls ahead of a close. Another's might be drafting job descriptions for a recruiting team. Those carry different risk, different reviewers, and different rollback stories, and one number covers both.

It cannot tell you whether the agent is in production. Using, in a press release, covers a sandbox that a platform lead enabled to see what the toggle does. It covers a pilot in one country with twenty users. We don't know how Workday defines using, and the release doesn't say, so we'd treat the count as a lower bound on curiosity rather than a measure of work completed.

And it cannot tell you what happened after the agent acted. How many runs, how many exceptions a person reviewed, how many outputs were reversed. Those are the questions your controller and your HR operations lead will ask in the first steering meeting, and no earnings release from any vendor answers them. Our agent readiness check starts from those questions for exactly this reason.

The 25 percent is a sales figure, and it will shape your renewal

The other number, AI driving more than 25 percent of new annual contract value, describes what customers bought rather than what they run. It is still useful, because it tells you how Workday's sales organisation is being measured. When a quarter of new contract value comes from AI, the account team's plan for your renewal will include agents whether your own roadmap does or not.

That is the moment to have a licensing read ready. Which agents are covered by what you already pay for, which are add-ons, and what the meter is if there is one. The release doesn't say, and we'd ask before the proposal lands rather than after. Our checklist for reading a licensing change works here even though nothing has been formally repackaged yet.

The adoption numbers to collect inside your own tenant

The adoption figures that matter are the ones no vendor can publish. For each agent you enable, we'd want runs per week, the share of runs a person changed or rejected, and the name of the person who reviewed them. If the Financial Audit Agent raised twelve flags in August and the controller closed nine as false positives, that is the adoption statistic for your tenant, and it decides whether the agent stays on in September.

This is also where the handoff gets written down. Which outputs an agent can commit on its own, which need a signature, and where that signature is recorded. Our piece on designing the handoff between human and agent work covers the design, and the Workday AI and agents hub tracks what the vendor ships next.

Ask for the number behind the number

Before your account team puts 5,500 on a slide, ask for a smaller number. How many customers in your industry, at your size, have run the specific agent you're being offered in production for two consecutive quarters. Three names who will take a call is a reference. The 5,500 on its own is a press release, and you can read the release yourself in five minutes.

Take the smaller question into the next meeting, and don't let the bigger number answer it.