Seven enterprise vendors reported through 28 August 2026 and put eighteen quantified agent claims into circulation. Nine are counts with no dollar attached. Five carry dollars, and every one of those is a non-GAAP operating metric management defines and no auditor tests. Not one reaches a reported financial line.
Workday, Okta, nCino, Best Buy, IFS, ServiceNow and Salesforce all reported in the ten weeks to 28 August 2026, and between them they put eighteen quantified or semi-quantified agent claims into circulation. Nine are counts of customers, deals, deployments or transactions with no dollar attached. Five are denominated in dollars, and every one of those five is a non-GAAP operating metric: annual recurring revenue, annual contract value, or a share of bookings. Not one appears as a line in an income statement or in a footnote reconciled to one. That is legal and it is normal. It is a problem this quarter because the same numbers are carrying an investment thesis and a multi-year platform decision.
The shape of the metrics is the part that outlasts the quarter. "Customers using one or more agents" has a floor of one and no ceiling, no usage threshold and no production requirement, so it rises whenever a new account tries anything and it cannot fall unless someone actively rips an agent out. It can go up in a quarter where every existing deployment shrinks. Workday's count went up 35 percent quarter over quarter across exactly the window in which the company was culling its own agent portfolio from about fifty down to about twenty, and the metric records consolidation and growth identically.
The gap is not evidence of dishonesty. It is a disclosure regime that has not caught up to a category being sold at platform scale. Until a vendor breaks out agent revenue as a reported line with a prior-period comparison, buyers and analysts are underwriting the category on management's arithmetic, and this report is the reconciliation nobody publishing the numbers has done.
Built from earnings releases, prepared commentary and trade coverage for seven companies reporting between 30 June and 28 August 2026. Each finding is labeled by evidence type, so a verified figure from a release is graded separately from a vendor's characterisation of its own business and separately again from my reading of the two together. Where the evidence cuts against the argument, it is printed rather than dropped.
Across seven companies, four claims are commentary with no number, nine are counts with no dollar value, and five are dollar-denominated. All five of those are non-GAAP operating metrics: ARR, ACV or a share of bookings. Not one of the eighteen appears in an income statement or in a footnote reconciled to one.
More than 5,500 customers using at least one organic agent, up over 35 percent sequentially, and AI drove more than 25 percent of new annual contract value. In the same release, total subscription revenue backlog was $27.403 billion, up 8.0 percent year over year. The 12-month backlog grew 14.2 percent. Both figures are correct and they measure different things, which is why quoting the first without the second misleads.
Workday's AI ARR approached $600 million, up more than 200 percent year over year. Against fiscal 2027 subscription revenue guidance of $9.94 billion to $9.95 billion, that is roughly six percent. Against total subscription backlog of $27.4 billion, about two percent. The growth rate is the headline; the share is the story, and only one of the two gets quoted.
Dozens of AI agent deals, several worth more than $1 million, and new products including Okta for AI Agents at roughly 30 percent of second-quarter bookings. Management said agent revenue is immaterial to the top line and will stay so this fiscal year. Revenue grew 11 percent to $805 million, cRPO 14 percent, total RPO 17 percent. The quarter was carried by core identity.
The release calls the company the platform for agentic banking and describes a dual workforce of AI agents alongside human teams. It contains no count of agents deployed, customers using agents, or revenue attributed to them. The one quantified AI reference is four enterprise customers holding more than $900 billion in combined assets renewing early with expanded AI commitments, which is a count of renewals.
Ask Blue launched as a conversational shopping and support assistant with no interaction volume, conversion effect, cost saving or headcount impact disclosed. The quarter was strong on its own terms: $9.78 billion of revenue, comparable sales up 4.1 percent, full-year guidance raised. Best Buy attributed the comp to computing and emerging categories. The trade coverage linking the two is adjacency, and Best Buy did not claim otherwise.
Aneel Bhusri, back as Workday's chief executive since February 2026, cut the internal agent count from about 50 to about 20, describing killing agents or rolling them into bigger ones, with roughly 15 new agents planned for 2026. A 60 percent withdrawal rate on a first-generation portfolio, given conversationally to the trade press. It is the best read on pilot-to-production conversion in the set, and the disclosure regime treats it as colour.
The Futurum Group's 1H 2026 survey of 830 IT decision-makers produced the 86.6 percent agentic-priority figure quoted in commentary on nCino's quarter. Futurum also ran the 664-respondent study behind IFS's agentic digital worker campaign, in partnership with IFS, and separately publishes earnings analysis on the same companies. Each role is ordinary alone. Reading the three together is what a procurement lead scoring a vendor deck should do.
A trade article dated 28 August 2026 reported revenue of $2.35 billion up 12.6 percent, subscription revenue of $2.169 billion and net income of $228 million. Those are Workday's fiscal 2026 second-quarter figures from a year earlier. Fiscal 2027 Q2 reported $2.649 billion and $2.471 billion. The agent metric in the same article is current, so the implied ratio of adoption to revenue is wrong by about 13 percent, in the direction that flatters the agents.
The report settles each metric against what the company actually wrote, then names the revenue line it should have touched and did not. Three words carry most of the weight in the headline number: "customers" is an account rather than a seat or a workflow, "organic" excludes anything that arrived through acquisition, and "use" has no stated threshold of frequency, volume or production status anywhere in the release.
Accounts with at least one non-acquired agent in any state of use. No usage threshold, no production requirement, no per-agent count. A 55,000-employee bank running one agent everywhere and a 200-person customer running one agent in one department each count as one.
Vendor-defined operating metric. Reaches subscription revenue only through the ACV it carries
Annualised recurring value of all AI products, agents included but not separable. Adaptive Decision Intelligence, Sana Enterprise and the agent portfolio all sit inside it. Workday's own filings describe company-reported AI ARR as an operating metric rather than GAAP revenue.
Subscription revenue over the contract term. About 6% of FY27 subscription guidance
Share of newly contracted annual value attributed to AI products in one quarter. The attribution rule is not disclosed, so an outsider cannot tell which products count or how a mixed deal is split.
12-month backlog, which grew 14.2%. Total backlog grew 8.0%
A count of closed deals containing an AI agent component, with the threshold undisclosed. "Dozens" and "several" are not auditable quantities. The most quoted Okta figure of the quarter, agent instances going from 50 to 1,500 in weeks, describes one unnamed customer environment over an unstated window.
cRPO, which grew 14%. Management states the contribution is immaterial
Annual contract value of AI products across the platform, and the only agent-adjacent metric in the set reported as a scale figure rather than a growth rate. Customers with agentic AI in production grew ninefold over nine months; first-time agentic buyers rose 45 percent year over year.
Subscription revenue of $3.877bn, up 24.5%. No attribution given, no reconciliation in the release
Annualised recurring value of one named product family, and the cleanest product-level figure in the set. The number that circulates is the combined Agentforce and Data 360 figure of roughly $3.4 billion, which includes about $1.1 billion of acquired Informatica Cloud ARR. That combined number is not an agent number.
Subscription and support revenue over the contract term
A count of renewals with expanded AI commitments, not of agents and not of AI dollars. Verifiable as an event, and it carries no adoption or revenue information. Subscription revenue grew 10 percent against non-GAAP operating income growth of 36 percent.
ACV guidance of $662.5m to $667.5m. No AI component disclosed
A product exists. Verifiable, and not a metric. The Q2 comparable sales gain of 4.1 percent predates the launch, and Best Buy attributed it to computing, Best Buy Business and emerging categories.
None claimed by the company
A share of an undisclosed transaction population reaching completion without escalation. No population, no period, and no stated method for determining end-to-end completion. It is also the only completion-rate metric published by any vendor in the set.
No public financial statement. Privately held, unverifiable by construction
The technology does not determine what the next two quarters reveal. The disclosure regime does, and disclosure practice changes discontinuously: one company moves and the rest follow within two cycles. Probabilities are subjective and are the least reliable content in the report. The earliest visible signs are the part worth monitoring, and each is observable from published documents within days of release.
At least one company in the set reports an agent or AI revenue line in a filing, with a prior-period comparison, by the end of the first half of 2027. Salesforce and ServiceNow are the likeliest first movers, since both already publish product-level dollar metrics and both have an incentive to make the number auditable while it is growing fastest. Peer pressure forces the rest within two quarters, the category becomes analysable, and the vendors whose agent revenue is small get repriced immediately.
Sign: a new line item or defined term in a 10-Q, or an investor-day slide reconciling a non-GAAP AI metric to a GAAP line
Adoption counts continue, definitions shift quietly between quarters, dollar figures stay inside non-GAAP operating metrics, and nobody reconciles anything. Commentary carries the story while the audited statements stay silent on agents. Buyers and analysts keep underwriting the category on management arithmetic, dispersion between vendors widens without becoming visible, and the correction arrives as a guidance cut rather than as a disclosure.
Sign: a metric's wording changing between two consecutive releases with no restatement of the prior period
A vendor that has been reporting an agent adoption count stops, or swaps it for a different metric, in a quarter where the original would have decelerated or fallen. This is the standard end-state for enterprise vanity metrics and it has happened before with seats, active users and consumption units. The withdrawal is more informative than any number the company published, and it resets how the whole set is read.
Sign: a metric present in three consecutive releases absent from the fourth, with no replacement and no explanation
The report also carries the two questions that separate a vendor measuring agents from one marketing them, namely how many agents have been withdrawn, merged or renamed in the last four quarters and what happened to the customers running them, and whether the vendor will state a production-usage threshold in your contract. It sets out seven leading indicators with where to watch each, a table of nine claims in circulation that no disclosure supports and what would settle each, separate implications for CIOs selecting multi-year agent platforms, software equity and credit analysts, procurement leads scoring vendor decks, and product leaders setting internal adoption KPIs, four questions it states plainly that it cannot answer without filing-level access, and a half-life on every figure in it. The call is falsifiable and the condition is printed: I would change position the first time a vendor in this set breaks out agent revenue as a reported line in a filing, with a prior-period comparison.
This isn't a vendor summary. Every sentence is labeled by what stands behind it: verified fact, vendor claim, third-party estimate, my assessment, hypothesis, or scenario. Sources are numbered and clickable. Forward-looking sections use scenarios with observable tripwires, not forecasts. It's the same method behind every market assessment I write.
Twenty-three pages, built from public sources with no client brief and no interviews. Read it in the browser or take the PDF.
Each report here answers a real question, directed and researched against public sources and evaluated against a stated assumption, then delivered as Word and PDF. If you're weighing a platform, sizing a category, or defending a number to a board, tell me the decision behind it and I'll tell you honestly whether a report is the right tool.
Commission an assessment