Salesforce closed its second fiscal quarter with 3.2 billion units of AI agent work billed in three months, a 97% jump from the prior quarter. Zoom posted its strongest enterprise growth in three years on the back of AI voice agents. Both events landed the same week, and three trade outlets covering them walked away with three different readings of what it means for how sales and revenue software gets priced.
The same numbers, three different stories
The trigger was Salesforce’s August 26 second-quarter fiscal 2027 report, followed a day later by Zoom’s own quarterly print. Salesforce disclosed that customers generated 3.2 billion Agentic Work Units (AWUs), its unit for a completed AI agent task, up 97% quarter over quarter, with 7 billion delivered since launch. Half of Agentforce bookings in the quarter came from customers refilling Flex Credits after burning through a prior allotment, a pattern Salesforce itself reads as proof that agent usage is becoming a real, recurring line item rather than a pilot.
CX Today framed it as a turning point for the whole licensing model: per-seat software, it argued, assumed a human sat behind every login, and that assumption breaks the moment an agent does the work instead. The piece pointed to Microsoft’s Dynamics 365 usage-based credits and Zoom’s shift toward consumption and outcome pricing for its newer AI products as evidence the seat is losing its place as the default unit of enterprise software.
Advertisement
Futurum Group read the same earnings print more cautiously. Its analysis called the Flex Credit refill behavior a genuine product-market-fit signal, but stopped short of declaring the pricing model settled, framing the real test as whether “stronger AI bookings and production usage translate into the organic revenue reacceleration management expects.” In other words: promising consumption data, unproven durability.
No Jitter’s coverage of Zoom’s quarter barely engaged the pricing question at all. It attributed Zoom’s 7.8% enterprise revenue growth, its best in three years, to platform consolidation and cross-selling among Phone, Contact Center and AI features, not to a change in how those features get billed. Where CX Today saw a pricing revolution, No Jitter saw ordinary account expansion with AI as one more attach.
Why the seat stopped being the natural unit
Per-seat pricing worked for two decades because it matched how software got used: one login, one salesperson, one predictable line on a budget. An AI agent breaks that match on both sides. It does not need a seat to update a CRM record, draft a follow-up email or score a lead, and it can do in an hour what took a rep a week, which means the old proxy for value (how many people are logged in) stops tracking the thing the vendor is actually selling (how much work got done). That is the mechanical reason all three vendors in this week’s coverage are running some version of usage-based billing alongside, not instead of, their seat licenses. Salesforce’s Flex Credits, Microsoft’s usage-based Dynamics 365 credits and Zoom’s outcome-priced AI features are three separate attempts at the same fix: find a unit that moves with agent output instead of human headcount.
Microsoft’s own numbers, as CX Today reported them, follow the same pattern as Salesforce’s: customer-service AI credit consumption rose fourfold quarter over quarter, a faster growth rate than the seat business it sits alongside. None of the three vendors has replaced seats outright. All three are layering a second, usage-tracking meter on top of the first, which is exactly why the coverage splits on whether that counts as “the end” of per-seat pricing or just an addition to it.
Where the coverage actually disagrees
That gap is the real finding. It is not that these outlets disagree on the facts. All three cite the same quarters, the same vendors and largely the same growth figures. They disagree on how much the pricing mechanism itself has changed. CX Today treats consumption and outcome pricing as the headline story, the thing that makes this quarter different from the last five. Futurum treats it as an interesting but still-unproven wrinkle inside a broader growth narrative. No Jitter barely treats it as a story at all, folding AI monetization into the same platform-expansion frame that has explained SaaS growth for a decade.
That range matters more than any single outlet’s verdict. It suggests the shift away from per-seat pricing is real at the vendor-disclosure level, Salesforce, Microsoft and Zoom are all now running blended models, but it has not yet become obvious enough at the buyer level for every outlet covering the same earnings calls to treat it as the central story. A change that clear would not produce three different frames from three publications reading the same numbers in the same week.
What it means for the RevOps leader
For a RevOps or sales-ops team, the practical consequence sits below the pricing-model debate: budgeting for CRM and revenue tooling is getting harder to forecast, not easier, right as AI features become mandatory rather than optional. A seat count is a number a buyer controls directly through headcount planning. A Flex Credit balance, an Agentic Work Unit tally, or a per-resolution fee moves with usage patterns that shift quarter to quarter and are, in Salesforce’s own telling, still accelerating at 97% sequential growth. Budget owners who modeled next year’s software spend on last year’s per-seat renewal are working from the wrong unit.
The fix is not resisting consumption pricing, the vendor data suggests it is arriving whether buyers like it or not, but building the same usage-monitoring discipline around AI credits that finance teams already apply to cloud compute spend: a named owner for consumption forecasting, alerts before a credit pool runs dry mid-quarter, and contract language that caps overage costs rather than let them float. Marc Benioff, Salesforce’s chair and CEO, told investors on the earnings call that the company is “turning AI into customer success at unprecedented scale,” and Robin Washington, Salesforce’s president and chief financial and operating officer, added that “AI is amplifying the power, reach, and value of our platform,” pointing to the strongest net new order value growth in four years as evidence the model is working for the vendor. Neither claim addresses what the same volatility looks like from the buyer’s side of the invoice.
SalesTech has flagged this tension before: an opinion piece in August argued seat-based pricing was actually winning the AI agent war, on the logic that buyers wanted predictable bills more than they wanted to pay only for agent output. This week’s numbers complicate that read rather than settle it. Salesforce, Microsoft and Zoom are not abandoning seats, they are running seat-based, consumption-based and outcome-based pricing in parallel, often for the same product line, which is its own kind of answer: nobody, including the vendors, is confident enough in one model to commit to it alone.
The same caution applies to the AI-agent ROI conversation more broadly. A recent SalesTech analysis of Salesforce’s own agentic AI leaders survey found that the fastest deployments were not the fastest to show return, a pattern that lines up with Futurum’s reluctance to call this quarter’s consumption growth durable. Fast usage growth and fast financial payoff are not the same claim, for vendors or for the buyers now trying to budget around both.
What to watch next
The next test is not another earnings call quoting bigger AWU or credit-refill numbers, vendors have every incentive to keep publishing those. It is whether a full renewal cycle, twelve months out, shows buyers actually accepting consumption or outcome terms at scale rather than negotiating back toward the predictability of a seat. Until then, treat “per-seat pricing is over” as a vendor-side observation still waiting on a buyer-side verdict, and budget accordingly.
At the negotiating table now, that means three concrete asks rather than a general objection to usage pricing: a cap on overage rates so a busy quarter cannot silently blow through the software budget, mid-cycle visibility into credit or AWU burn rather than a surprise at renewal, and the right to true-down as well as true-up if agent adoption stalls after launch. Vendors that are still running seat and consumption pricing in parallel, which is every vendor in this week’s coverage, have room to grant those terms without giving up the upside they are chasing. The RevOps teams that ask now, while the models are still being built, will get better terms than the ones that wait for the next renewal to find out what a bad quarter costs under a meter they do not control.
Source: Salesforce
