Reading the Agent ARR: What $1.5bn Does and Does Not Say

What Salesforce reported

On 26 August 2026 Salesforce published its second-quarter fiscal 2027 results with a line that agent-watchers will have noted: "Agentforce ARR exceeded $1.5 billion, up over 240% Y/Y"¹. The next sentence carries a condition: "Effective Q2 FY27, Agentforce ARR includes our AI offerings, Slackbot and Headless 360"¹.

This brief takes those two sentences as its subject. It is not a verdict on whether Agentforce works. It sets out what the number measures, what sits beside it in the same release and earnings call, and what a decision-maker who has to budget for agent spend should ask before relying on it. Every metric below is defined and reported by Salesforce unless the text says otherwise.

The figures, and who vouches for them

ARR and AWUs are management metrics. The revenue lines are accounting figures. They answer different questions and should not be read against each other as if they were one series.

The Agentforce Apps line is also not a measure of agent revenue. Salesforce describes it as Agentforce Sales, Service, Marketing and Commerce, plus Flex Credits and Slack¹, which is largely the established applications business. Its 8% growth tells the reader how fast the segment carrying the agent products is growing. It does not say how much of that segment is agent-derived, and the release does not break that out.

Scale matters here. Annualising the quarter's subscription and support revenue gives about $43bn (my calculation, $10,820m multiplied by four), so an Agentforce ARR of just over $1.5bn is roughly 3.5% of it. Even very fast growth on a base that small moves aggregate growth very little. That arithmetic explains why a 240% headline and single-digit segment growth can coexist without either being wrong. It does not make the 240% comparable to anything.

Excluding Informatica, subscription and support growth is about 7.1% if the full $440m contribution is acquired revenue absent from the prior-year base (my calculation; the release states the contribution but not that assumption)¹.

What the ARR figure does not establish

Salesforce defines Agentforce and Data 360 ARR as "the annualized recurring value of active Data 360 and certain generative artificial intelligence ('AI') subscription agreements, including those for Agentforce, generative AI products and features, and Headless 360, that were executed at the end of the reporting period"¹. The release adds that it may "update this metric to incorporate new products, feature offerings, or acquired technologies that meet its definition", and that "Beginning in Q2 FY27, Agentforce ARR includes our AI offerings Slackbot and Headless 360"¹.

Three consequences follow.

• It is contracted value at a date, not revenue and not usage. Secondary coverage blurs this quickly: Futurum's note says Agentforce "is on track for roughly $1.5 billion in revenue this year"³, which the release does not say.
• The growth rate is not shown to be like-for-like. We found no statement in the release on whether the prior-year comparison for "over 240%" was recast to the new scope. Ivris Tech reaches the same reading: the Q2 materials "do not state that the historical Agentforce ARR comparator was recast to the same scope"⁴. Gene Dai adds that the release "does not provide a bridge" between like-for-like growth and the expanded definition⁵. Ivris puts the Q1 figure at $1.2bn⁴; we have not checked that against Salesforce's Q1 release.
• The change was disclosed. Ivris is explicit that this "is not evidence that Salesforce hid or manipulated the metric"⁴. The problem is comparability, not concealment.

These two commentators are independent of Salesforce but are not replication. Both work from the same release. Ivris Tech is a B2B marketing and RevOps services firm; Dai is a co-founder of an AI company. Neither has access to Salesforce's contract data.

Salesforce disclosed the scope change. What it has not published, in the materials examined, is a like-for-like series that would let a buyer separate product growth from definition growth.

The refill signal

On the call, Miguel Milano, Salesforce's President and COO, said bookings "grew triple digit" and that "50% of the bookings came from customers refilling the tank"². He described the mechanism as customers who "consume, they use the Flex Credits, they want more, they raise their hand"². The 50% is a share of the quarter's bookings, not of ARR, and no base is disclosed. The transcript is a Motley Fool page that states it was partly created with language models and may contain errors, so quotations here should be checked against the recording before external use.

The figure supports two readings that the evidence cannot separate. On the value reading, customers find agents useful, burn through their credits and buy more. On the depletion reading, credits are allocated or priced so that they run out as a matter of course, and refills are the normal mechanics of a prepaid or pay-as-you-go model. Milano's own account leans on the second as well as the first: customers "buy as they go", Salesforce sometimes charges "for overages", and "many times they buy up front a bunch of credits"². Under that description, a refill records that credits were consumed. It does not record whether the work done with them was accepted by the business.

Futurum repeats the figure as bookings "tied to customers refilling the tank after consuming credits"³. That is a restatement, not a second source. Futurum discloses that it has done "research, analysis, and advisory services with many technology companies, including those mentioned in this article", and its site lists a case study on Agentforce completed in partnership with Salesforce³. It should be read as an interested party.

Credits are being resized under the number

In September Salesforce restructured its Sales and Service editions. According to CIO, Core rose from $175 to $195 per user per month and now includes 500,000 Flex Credits; Advanced is $395, with 1 million credits, replacing a $350 edition; and Max stays at $550 but now includes 2.75 million credits instead of 1 million⁶. Salesforce says the new editions deliver "from 50% to 70% greater value" than those they replace⁶.

Analysts quoted by CIO are less certain. Pareekh Jain: a package "may offer 60% more theoretical value, but that benefit can disappear if much of the bundled functionality or Flex Credits goes unused"⁶. Manoj Chandra Jha says bundling "reduces the line-item visibility CIOs previously relied on", and that most enterprise finance teams are still learning to forecast credit consumption⁶. In a May piece, Ashish Chaturvedi of HFS Research predicted that customers "will throttle agent usage to control costs"⁷. That is a forecast, not an observation.

The inference, which is mine and not a reported fact, is that Q2 refill behaviour was observed under the old allocations. Larger bundled allocations change when customers reach the ceiling, so the Q2 refill share is a weak guide to future bookings and an equally weak guide to what a buyer's credit spend will look like.

What is proven, and what is not

What these sources support: Salesforce widened the scope of its headline agent metric in the quarter it reported over 240% growth, and disclosed it. The disclosed revenue lines show segment growth of 8% in constant currency for Agentforce Apps and 11% for total subscription and support. Half of Agentforce bookings came, by Salesforce's account, from customers who had already used credits and wanted more.

What they do not support:

• That agents lack value. Nothing here measures outcomes in either direction.
• That refills mean unmet demand rather than credit depletion. The two cannot be separated from the disclosed data.
• Any unit economics. Salesforce has not disclosed the cost of an agent task, or credit consumption per customer.
• That AWUs measure delivered value. Salesforce defines an AWU as "a measure of discrete tasks executed by AI agents in production", with examples such as resolving cases, updating records or triggering workflows¹. Dai's reading is that "an AWU identifies activity performed by an agent", whereas "an accepted outcome says the business kept the result"⁵.

Also unresolved: the contribution of Slackbot and Headless 360 to the ARR figure; the split between paid and pilot ARR; and the number of paying customers in total. The call gives growth rates for accounts with agents in production and for customers added to production, but no absolute counts². The 10-Q was not examined for this brief, so any further definitional detail there is unchecked.

What a decision-maker should conclude

On its own, the headline figure is evidence of what Salesforce is selling, not of what customers are getting. For anyone setting a portfolio position on agent platforms, three tests follow.

1. Ask for the definition, its version date and its comparability status with any vendor agent number. If the vendor cannot say whether the comparison base was recast, treat the growth rate as unquantified.
2. Ask for consumption against entitlement in your own estate: the share of included credits used, expiry terms, the rate card, the price of a refill, and a plan for twice the forecast consumption. The analysts quoted here agree that finance teams are not yet good at this forecast⁶.
3. Separate activity from outcome. None of the disclosed measures (ARR, AWUs, bookings) records accepted work, so track accepted outcomes per credit yourself before treating usage growth as value.

Until Salesforce publishes a like-for-like series and credit consumption data, read agent ARR as an indicator of sales scope and buy against your own measured outcomes.