On the evening of August 5, Figma reported the fastest revenue growth in its history. Sales rose 48 percent to $370.1 million, ahead of every forecast, and the company raised its full-year guidance. The stock fell 18 percent. The next day Datadog, whose business is billed by usage, sank further after its largest AI customer, which analysts estimate to be OpenAI, cut consumption starting in June. A week later Workday jumped nearly 18 percent, its biggest one-day gain in a decade, after Reuters reported that Silver Lake had been in talks for months to take the company private in a deal near $51 billion.

Three companies, three directions, one month. Software has spent 2026 inside what Wall Street calls the SaaSpocalypse debate, swinging between the fear that AI will destroy its pricing and the hope that it will not. The sector's exchange-traded fund fell more than 2 percent on that Thursday and, even after climbing more than 35 percent from its April low, remained down for the year. But the three stocks share something more specific than the sector's mood. Each company is moving its product off the per-seat model and onto a meter that charges for the work the software does. That is the real change, and the market has not yet learned to read it.

Per-seat pricing was never about measuring use

The seat was never a meter. For roughly three decades, enterprise software was sold by the head: so much per user per month, scaled to the employee count. The model's genius was not that it measured anything. Unused seats were routine, and both sides knew it. The genius was that the seat anchored the software bill to a number the customer already owned. The CFO had a headcount, and software became a fixed, predictable line item that grew on its own schedule, because companies added people and the bill rose with them. The waste of idle licenses was the price of certainty, and buyers paid it for decades, because certainty was what the contract was for.

The analyst firms now date the seat's end. IDC says pure seat-based pricing will be obsolete by 2028, with most vendors forced to refactor how they sell as agents take over repetitive tasks. Gartner projects that at least 40 percent of enterprise SaaS spending will shift to usage-, agent-, or outcome-based pricing by 2030. The reason is that the anchor is gone. When an AI agent completes a task, the unit of value stops being the employee who used to do it, and the one number the CFO already had no longer maps to software value. The vendor is left holding a price with no unit underneath it.

Workday, GitLab, and Figma each chose a meter

Workday, whose software runs payroll and finance for more than 11,000 organizations, went first. At its customer conference in September 2025 it launched Flex Credits, a single credit type covering its AI agents, its data cloud, and its knowledge platform. A credit is consumed when an agent completes an action, not when a query is made or a token is spent, and the company publishes a rate card that ties credits to recognizable activities such as recruiting and contract negotiation, so a customer can compute a cost per process. The credits are sold as an annual subscription allotment, to preserve budget predictability. Then-CEO Carl Eschenbach told analysts the company was "focused not on just seats, but actually revenue per seat."

GitLab went hybrid. Executives said in late 2025 that the company was moving from a pure seat business to a mix of seats and usage, and the finance chief said the seat-based revenue retention metric would become less relevant as usage revenue grew. Usage is sold as credits with an on-demand price of $1 each, and Premium and Ultimate subscribers receive monthly credit allowances per user. In June 2026 GitLab went further, folding seats and credits into one annual dollar commitment called GitLab Flex, which customers can reshape month to month without new contracts. The company's own announcement states the case: "Most enterprise software contracts lock-in spend before teams know how they will use it."

Figma kept the seat and metered on top of it. It restructured seats by role in 2025 so that designers pay more than developers and viewers, then in March 2026 began charging for AI features with monthly credits: each user gets a personal allowance that resets monthly and never rolls over, and beyond that sit a shared subscription pool and pay-as-you-go billing under a cap the administrator sets. In its first full quarter selling AI by credit, revenue growth accelerated to 48 percent and more than 80 percent of its largest customers used credits weekly.

The meter moved the risk of non-use onto the vendor

Here is what changed when the unit went from the worker to the work. Under seats, the customer paid for capacity whether or not it was used, and the vendor collected its billings on contract and on schedule. Under credits, the vendor's revenue depends on the customer's agents actually getting used, every month, indefinitely. The risk of non-use, for decades the buyer's problem in the form of idle licenses, is now the vendor's problem in the form of idle meters. No one has a historical series for agent utilization, so every revenue model embeds a guess.

Figma is the live demonstration. Its AI features cost real money to run, and while they are in beta the company does not charge for them, so it absorbs the inference cost with no offsetting revenue and warns investors to expect gross margin to vary from quarter to quarter. Research and development more than doubled, total operating expenses nearly doubled, and the company swung to a quarterly operating loss of more than $117 million while its adjusted margin fell from 16 percent to 10 percent. Its finance chief added a detail that cuts to the core of the seat model: the company is "hiring fewer people than we originally had planned," because AI raises staff productivity. The anchor is breaking inside the vendor itself.

Datadog ran the same logic through the stock tape. Its revenue is metered by usage, and its shares fell 19 percent in a day, the worst drop since its 2019 listing, after its largest AI customer, which analysts estimate to be OpenAI, cut usage since June. The meter works until the client dials it down, and then the meter is the bad news. The sector spent decades teaching the market to price certainty, in billings and contracts. It is now asking the market to price utilization, and this month's tape is what a market looks like when it must price something it has never priced before.

The credits are a bid to buy certainty back

The vendors know the problem, which is why every meter they built is wrapped in the old model. Workday sells credits as an annual allotment, alerts customers at 80, 90, and 100 percent of balance, and reconciles overages with account teams rather than cutting service. GitLab Flex is one annual dollar commitment reshaped monthly. Figma's credits are personal and monthly, a seat in everything but name, with spending caps where the seat used to be. The credit systems are not a clean break from per-seat pricing. They are per-seat pricing's attempt to survive contact with the work.

Buyers have their own objections, and they deserve the strongest statement. Sanchit Vir Gogia of Greyhound Research argues that predictable licensing is giving way to vague units such as credits, interactions, and events, with vendors shifting cost volatility and overconsumption risk onto customers. Analyst Adam Mansfield warns of buyers who exceeded conversation thresholds on Salesforce's Agentforce without anyone having defined what counted as a conversation, owing sums that careful contract language could have avoided. The sellers' case is just as direct: price should track value, and a credit for a completed action sits closer to value than a credit for a body. This analysis takes no position on which case wins. The projections themselves disagree on timing, IDC saying 2028 and Gartner 2030, and both may be wrong in either direction.

The market is still paying for certainty, not for the meter

The August tape, read as a whole, says something narrower than "AI kills software." The companies the market rewarded were the ones that kept the most seat-like structure. Workday drew buyout interest at a price Reuters Breakingviews modeled at roughly five times estimated fiscal 2027 revenue, and Zacks' Brian Mulberry called the interest potentially the strongest evidence yet that the AI-driven pessimism on software had overshot. Constellation Research's Ray Wang put the bull case plainly: "You've got data, you've got distribution, you're still gonna win." The talks are unconfirmed by either company and may produce nothing, but the premium on the fixed-base model is the fact that matters. Atlassian, still sold largely by the seat, delivered its most profitable quarter since 2021 and saw its shares rise 35 percent in a day, the biggest gain since its 2015 listing, and its chief executive said he would buy up to $250 million of stock. The companies the market sold were the meters: Figma, Datadog, HubSpot. Even the cautionary tale ran that way, with Airtable sold to Bending Spoons for less than $1.3 billion, about a tenth of its 2021 valuation.

The stock market prices the anchor. The product roadmaps are removing it, one release at a time. The gap between the two is this year's volatility, and it will stay the volatility until the credits prove themselves.

The meter is only as good as the proof it produces

The transition will land where the credits land, and the credits are a wager on proof. Under the seat, the renewal conversation asked how many people the software covered. Under the meter, it asks what the software did, and the vendor's own meter is its invoice for its claimed output. Workday prices the completed task, GitLab prices the agentic request, Figma prices the AI feature, and each will have to demonstrate, quarter after quarter, that the work its agents complete was worth what the meter charged for it. Vendors that can show the work will keep their rates. Vendors whose meters bill for work the customer cannot see will find their credits discounted at renewal, which is how the seat died too, one budget review at a time. The seat's real product was certainty, and certainty was never the meter's business; someone still has to supply it, and the companies the market is paying for this month are the ones that do.

Primary sources

  1. Workday's blog post on its journey to usage-based AI pricing for the Flex Credits mechanics, and CIO.com's report on new software pricing metrics for the IDC projection and the views of Greyhound Research's Sanchit Vir Gogia and analyst Adam Mansfield.
  2. GitLab's June 10, 2026 press release for the GitLab Flex commitment structure and the monthly reshaping feature, and Figma's blog and AI Credit Terms for the March 2026 credit system.
  3. The Next Web's report on Figma's second-quarter results for the revenue, operating loss, margin swing, and finance chief comments, and Yahoo Finance and Digital Today for the August 6 sell-off figures, the Datadog and HubSpot moves, and Atlassian's results.
  4. UC Today and Reuters reporting on the Silver Lake-Workday talks, including Reuters Breakingviews' valuation model, diginomica's discussion of the Gartner projection, and Yahoo Finance's report for Constellation Research's Ray Wang.