In the first quarter of 2026, something happened to enterprise software stocks that had never happened before: the forward earnings multiple on the average listed software company fell below the multiple on the S&P 500 itself. Not converged with it. Fell beneath it. A sector that spent the 2020–2022 run trading at roughly four times the market’s multiple was, by March, being priced as a discount to it. The iShares Expanded Tech-Software ETF was down more than a fifth for the year. Something in the order of two trillion dollars in market value had gone missing since the previous September’s peak.
The instinctive explanation, repeated in enough boardrooms and earnings calls to have hardened into consensus, is that AI agents are eating the seat. Salesforce, Workday, Atlassian, Monday.com: all built revenue models on charging per logged-in human, and now a plausible story exists in which a fraction of those humans get replaced by software that does not need a login. Add a corporate hiring freeze that stalls the natural growth engine of seat expansion, and a wall of incremental IT budget now flowing toward OpenAI and Anthropic instead of the incumbents, and the multiple compression looks not just explicable but overdue.
It is a clean story. It is also, on the evidence, an incomplete one, and the part it leaves out is more useful to a CEO than the part everyone has already priced in.
The company that should be dying isn’t
Start with ServiceNow, because it is the control experiment the “AI kills seats” thesis cannot easily explain away. It sells almost entirely into enterprise IT and HR workflows, the exact terrain agentic automation is supposed to hollow out first. And yet in its most recent quarter, ServiceNow’s AI-related annual contract value crossed the billion-dollar mark, growing more than 40 percent quarter over quarter, with management guiding toward $1.5 billion by year-end and a stated ambition for AI to represent 30 percent of total contract value by 2030. Half of its net new business is now non-seat, priced instead on workload and consumption. Customers running its agentic tools in live production grew ninefold in nine months. Subscription revenue is growing at roughly 24.5 percent, comfortably ahead of Workday and close to double Salesforce’s pace, even as gross margin compresses a couple of points to absorb the hyperscaler compute costs that come with actually running the agents rather than just marketing them.
If AI were simply substituting for headcount, and headcount were simply the thing SaaS vendors monetised, ServiceNow’s own seat-adjacent business ought to be shrinking under the same pressure crushing its peers. It isn’t. Something other than “agents replace people” is doing the sorting.
What Salesforce’s pricing chaos actually reveals
The more instructive case sits one rung over, at Salesforce, whose Agentforce pricing has been rewritten three times in under two years and still hasn’t settled. It launched at two dollars per conversation, a metric so poorly specified that customers couldn’t predict their own bills; adoption stalled at roughly eight thousand of Salesforce’s more than 150,000 customers. The company pivoted to “Flex Credits,” ten cents per discrete action, sold in blocks of a hundred thousand for $500 — an improvement in granularity that still left a five-person support desk handling normal daily volume looking at bills approaching $20,000 a month, with no guaranteed relationship between spend and result. Only in late 2025 did Salesforce fall back on something familiar: flat per-user licensing, $125 to $550 a month, bundling the AI back into a subscription nobody had to reason about.
Compare that scramble with Intercom’s AI product, Fin, which from early on charged a flat 99 cents per resolved conversation. One metric. Success defined at the point of sale. No ambiguity about what a customer was buying. The difference between Salesforce’s three pricing rewrites and Intercom’s single stable one is not a difference in AI capability. Both companies have models that can hold a customer service conversation. It is a difference in whether the vendor already owned a workflow specific and countable enough to define what “done” looks like.
That is the piece of the story the market has been pricing without fully articulating it: this is not primarily a contest between AI-native challengers and legacy seat-sellers. It is a sorting of every software company, AI-enabled or not, according to whether it can produce a believable, auditable unit of value for what it sells. For decades, “per seat” was a serviceable proxy for value nobody could measure directly — you couldn’t easily prove what a CRM login was worth to a sales rep’s quarter, so you priced the login and let renewal rates stand in for satisfaction. Agentic AI didn’t create that ambiguity. It exposed it, by making measurable, outcome-priced alternatives suddenly available for comparison. Once one vendor in a category can tell a CFO “you pay $0.99 per problem actually solved,” every competitor still charging for access alone looks, for the first time, like it might be charging for nothing in particular.
The accounting trap nobody is pricing in yet
Here the second-order consequence gets interesting, and it belongs on the desk of every CFO now negotiating an outcome-based vendor contract or trying to design one to sell. Deloitte’s technical accounting guidance on this shift, published earlier this year, spends most of its length on a single, unglamorous question: when a company charges per successful outcome, is it promising continuous access to a capability (a “stand-ready obligation,” recognised over time) or a defined quantity of successful results (recognised as each one is delivered)? The answer changes when revenue shows up on the income statement, sometimes by entire quarters, and it depends on granular contract mechanics — whether unused credits roll over, whether a fixed minimum applies, whether hitting a volume threshold ends the arrangement early. Layered on top are questions about whether letting a vendor use your data to retrain its model counts as non-cash consideration owed back to you.
None of this is theoretical. Two companies delivering functionally identical AI outcomes, on economically similar contracts, can legitimately report very different revenue growth purely because their finance teams made different calls about contract structure. Boards and investors are not yet equipped to tell the difference, which means the next several quarters of “AI revenue” headlines deserve more scrutiny, not less: the accounting is unsettled enough to flatter almost any framing a management team chooses.
Where this leaves Indian software, on both sides of the contract
The repricing is not a Silicon Valley story with India as a footnote. Freshworks, still substantially built and engineered out of Chennai, posted its second consecutive profitable quarter in mid-2026 on 16 to 18 percent revenue growth, even as it navigates the same question every one of its larger American peers is wrestling with: how much of its support and IT service management business can it defensibly reprice around resolution rather than seats before its own customers revolt the way Salesforce’s did. Gurugram-based Leena AI offers the sharper data point on the other side of that bet: it moved early from consumption pricing to outcome-based billing tied to measurable HR and IT resolution, and reported that revenue accelerated once customers had budget clarity rather than a metered bill. An AI-native vendor with no installed base of legacy seat contracts to defend has less to lose, and arguably more to gain, from making that jump before an incumbent is forced into it.
The more consequential exposure sits with the Indian enterprises and GCCs on the buying side of these contracts, who are about to discover that procurement teams built to negotiate per-seat SaaS renewals are not built to audit an outcome-based bill. A vendor invoice priced on “successful resolutions” is only as trustworthy as the definition of success the contract specifies, negotiated once, usually before anyone has seen a year of real data. The CEOs and CFOs signing those contracts now are setting a template their finance teams will live with for years.
What actually needs to change in the boardroom
The useful question for a CEO is not “how many of our seats will AI eliminate.” It is whether the business already owns a workflow narrow and countable enough to define, sell, and audit an outcome — and if it doesn’t, whether pretending otherwise in a hastily redesigned pricing page will do more damage than staying on subscription honestly priced. It is also, less comfortably, a question for finance: whether the revenue recognition policy for the next generation of contracts is being decided deliberately, by people who understand the stand-ready-obligation test, or by whoever in product happened to ship the pricing page first. The companies that get bracketed with the multiple-compression story over the next two years may turn out to be sorted less by how much AI they deployed than by how precisely they could ever say what they were charging for.
