Salesforce Is Selling the AI That Could Eat Its Own Per-Seat Business

Begin with the model, because the model is the whole story. Software-as-a-service — the business model that produced a generation of the market's best-performing stocks — is built on per-seat pricing. A company like Salesforce sells its customer-relationship-management software to an enterprise and charges a recurring monthly fee for each named user: every salesperson, every support agent, every marketer who logs in is a "seat," and the enterprise pays for all of them. The magic of this model is that it grows almost automatically: as a customer's business grows and it hires more people, it buys more seats, so Salesforce's revenue from that customer rises without Salesforce doing anything. Land a customer, then expand as they add employees. It is recurring, it is predictable, it compounds, and it made Salesforce one of the largest software companies on Earth.

Now consider what an AI agent does. Agentforce, Salesforce's flagship AI product, sells autonomous agents that can handle customer-service inquiries, qualify sales leads, and perform the tasks that human employees with Salesforce seats used to perform. The explicit value proposition — the reason a customer buys it — is that the AI agent can do the work of a human, which means the enterprise needs fewer humans to do that work. And fewer humans doing the work means fewer human seats the enterprise needs to license. The product Salesforce is most excited to sell is the product that shrinks the headcount its own pricing model depends on. That is the dilemma, and it is not a hypothetical: Salesforce's own customers are reportedly already seeing seat reductions on the order of 10% as AI makes their service agents more efficient, and the broader software industry is showing the first signs of the same compression.

This essay is about that collision — between the AI revenue Salesforce is racing to build and the seat revenue it risks eroding — and about why a strong company doing the strategically correct thing is nonetheless facing a genuinely uncertain future that its 33% stock decline reflects rather than overstates.

The race between the new dollars and the old ones

The central question for Salesforce is brutally simple to state and genuinely hard to answer: is the new AI revenue growing fast enough to outrun the erosion of the old seat revenue? Frame the two numbers honestly. Agentforce annual recurring revenue has crossed $1 billion, growing more than 200% year over year — an explosive, impressive ramp, and a genuine achievement. But Salesforce's total revenue runs over $40 billion a year, the overwhelming majority of it the traditional per-seat subscriptions. So the AI business, for all its triple-digit growth, is still only a small fraction of the whole — and the whole is exposed to the seat compression the AI is causing.

Do the arithmetic of the race. If Salesforce's enormous legacy seat base shrinks even a few percent a year as customers replace human workers with agents, that is a drag of well over a billion dollars annually on the biggest part of the business — and the new Agentforce revenue, impressive as its growth rate is, starts from a small base and has to grow enormously just to offset that drag, let alone push total growth higher. This is why Salesforce's overall revenue growth has decelerated to the low teens and why management's guidance, even amid the Agentforce excitement, has been soft enough to spook investors: the market has done this subtraction and concluded that the new engine, while real, is not yet clearly large enough or fast enough to overcome the headwind the new engine itself creates in the legacy business. A 200%-growing billion-dollar product is a wonderful thing — but when it is both the company's best hope and the cause of its biggest risk, the net effect on total revenue is the only number that matters, and that number is, for now, merely good, not great.

The pricing chaos is the tell

If you want evidence that Salesforce has not yet figured out the economics of its own AI transition, look at how many times it has changed its mind about how to charge for Agentforce — because a company that keeps re-engineering the price of its flagship product is a company still searching for whether and how it makes money. Salesforce first priced Agentforce on a consumption basis, charging roughly two dollars per conversation the AI handled, a model that aligns the price with the work done. But consumption-based revenue is harder to predict and harder to model than subscriptions, Salesforce's investor base is built around subscription metrics, and the early adoption was tepid — by some accounts, of the first several thousand Agentforce deals signed, only a portion were actually paying.

So Salesforce pivoted. It introduced per-user licenses at $125-plus per month through a new "Agentic Enterprise License Agreement," bringing seats back as the primary wrapper but now reframing them as a "digital workforce" — charging per agent rather than (or in addition to) per human. It now runs multiple pricing models for Agentforce simultaneously. There is a reasonable argument, which some make, that experimenting with several pricing approaches is the right thing to do in a genuinely new market. But there is also a less flattering reading: a company that has cycled from consumption to seats to a hybrid in the space of a couple of years, for the product on which its entire future supposedly rests, is a company that does not yet know what that product is worth or how to capture its value. The pricing confusion is not a detail; it is a window into how unresolved the fundamental economics of the AI-agent transition remain, beneath the confident $1.2 billion ARR headline.

It is not just Salesforce — the whole model is cracking

The most important evidence that this is a structural threat rather than a Salesforce-specific worry comes from looking across the software sector, because the per-seat model is cracking everywhere at once. The clearest warning came from Atlassian, the maker of workplace-collaboration software, whose enterprise seat counts fell for the first time in the company's history — a milestone that sent its stock down sharply, because a seat-based software company whose seats are no longer growing is a growth company that has lost its growth engine. HubSpot and Monday.com, two other per-seat software leaders, saw their stocks fall roughly 50% and 44% over a comparable period. The entire category of software priced per human user is being repriced by a market that has suddenly grasped what AI agents mean for the headcount those seats represent — a repricing so violent that one stretch in early 2026 wiped out something on the order of $285 billion of software market value in a matter of weeks, as investors marked down per-seat business models en masse on the seat-compression threat.

This sector-wide repricing matters for two reasons. First, it confirms that Salesforce's seat-compression risk is real and general, not a passing scare — when an entire industry's stocks fall in unison on the same fear, the fear usually has substance. Second, it means Salesforce is navigating its transition not in isolation but amid a broader collapse in confidence in the business model it shares with its peers, which raises the stakes and narrows its margin for error. The decade-long bull market in software stocks was built substantially on the near-magical reliability of seat expansion. AI agents have introduced, for the first time, a credible mechanism by which seats can contract — and the market is rerating the whole sector to reflect that the magic may be ending.

What the bulls genuinely get right

In fairness, the bull case is serious and Salesforce's strategy is, in important respects, exactly right — the critique is about the difficulty of the transition, not the wisdom of attempting it. Several points genuinely favor Salesforce. First and most important, Salesforce is doing the strategically correct thing by cannibalizing itself: the AI-agent disruption is coming whether Salesforce participates or not, and far better to be the company selling the agents than the company being disrupted by a startup that sells them instead. A leader that leans into the technology threatening it, rather than denying it, is a leader giving itself the best chance to capture the new value. Second, the new value could genuinely exceed the old: an AI agent that does the work of three human employees could plausibly be priced higher than three seats, because the customer is buying an outcome — labor — rather than a login, and labor is worth far more than software access. If Salesforce successfully prices its agents as digital workers commanding digital-worker wages, AI revenue could more than replace seat revenue. Third, Salesforce owns the data and the customer relationships and the workflow that make its agents uniquely capable — an agent is only as good as the data it can act on, and Salesforce sits on the definitive system of record for its customers' sales and service operations. And fourth, the Agentforce growth is genuinely real and fast, and the stock's 33% decline means a great deal of the seat-compression fear is already priced in.

The honest synthesis is that Salesforce is making the right bet in a genuinely hard situation, and the outcome is genuinely uncertain. The bull case — that AI-agent revenue, priced as digital labor, more than replaces the eroding seats — is plausible and may well prove correct. The bear case — that the legacy seat base compresses faster than the AI revenue can scale, and that Salesforce spends years with decelerating total growth while it sorts out pricing and economics — is equally plausible. What is not credible is the pre-AI assumption, which the stock once embodied, that Salesforce's seat-expansion machine would simply keep compounding forever. That machine is now in question, by Salesforce's own hand, and the honest position is that no one — not the bulls, not the bears, not Salesforce's own repeatedly-changing pricing team — yet knows how the math nets out.

Cheap, or a value trap?

The 33% decline raises the question every falling former-darling provokes: is Salesforce now cheap — a great company on sale — or a value trap, a stock that looks inexpensive precisely because its growth model is breaking? The bull points out that after the drop, Salesforce trades at a far more modest multiple than the premium it commanded for years, generates enormous free cash flow, is highly profitable, and is buying back stock — the classic profile of a quality business available at a discount. On those metrics, it looks like value.

But the value-trap risk is real and specific, and it is the same trap that has caught many decelerating software leaders. A stock that has de-rated from a growth multiple to a value multiple can keep de-rating if the growth keeps slowing, because the market re-prices not just the current earnings but the expected trajectory — and a company transitioning from reliable seat expansion to an uncertain agent-revenue race has a trajectory that is genuinely harder to forecast and easier to disappoint. The danger is that Salesforce looks cheap on the old framework (a profitable, cash-generative software giant at a reasonable multiple) while the new framework — a company whose core pricing model is under structural threat and whose replacement model is unproven — justifies an even lower multiple still. Cheap relative to its history is not the same as cheap relative to its future, and when the future involves dismantling the engine that created the history, the historical multiple is a poor guide. The honest answer is that whether Salesforce is cheap depends entirely on the unanswered question at the heart of this essay — whether agent revenue outruns seat erosion — and a stock whose cheapness is contingent on an unresolved existential question is not obviously cheap at all. It is a bet, priced as if it might be a bargain.

This is compounded by a subtler discomfort for Salesforce's investor base: the shift from subscription to consumption-influenced economics changes the very nature of the stock. Investors paid a premium for SaaS precisely because subscription revenue was so predictable — the recurring, contracted, high-visibility revenue that let analysts model years ahead with confidence. AI-agent revenue, tied to usage and outcomes, is inherently less predictable, lumpier, and harder to forecast. So even in the optimistic scenario where agent revenue grows beautifully, Salesforce may be trading a high-multiple-deserving predictable business for a lower-multiple less-predictable one — another way in which the AI transition, even if it succeeds on revenue, may compress the valuation the company once commanded. The model that made the stock a premium compounder is being replaced by one the market has historically paid less for, regardless of how the revenue race resolves.

The deeper irony

There is a deeper irony worth sitting with, because it captures something true about this entire moment in software. Salesforce's marketing for Agentforce leans heavily on the promise that AI agents will let companies do more with fewer people — that the digital workforce will reduce the need for human headcount, cutting costs and boosting productivity. This is a genuinely compelling pitch to a customer's chief financial officer. But every customer who believes that pitch and acts on it is a customer who will, logically, reduce the human headcount that drives their Salesforce seat count. Salesforce is, in effect, running a marketing campaign whose central promise is "buy this and you will need fewer of the seats we sell you." The more successful Agentforce is at delivering on its core value proposition — replacing human labor — the more it erodes the human-seat base that is still the overwhelming majority of Salesforce's revenue.

This is the innovator's dilemma in its purest, most self-aware form. Salesforce cannot decline to build Agentforce, because then a competitor would, and Salesforce would be disrupted from outside. But by building it, Salesforce accelerates the erosion of its own model from inside. The only winning move is to ensure that the new agent economics generate more revenue than the old seat economics lose — and that is precisely the race whose outcome is unknown, whose pricing keeps changing, and whose uncertainty the falling stock price honestly reflects. There is no version of the future in which Salesforce simply keeps selling ever-more human seats; that world is ending, and Salesforce is helping to end it, in the hope of owning the world that replaces it.

The kicker

Salesforce is a great company facing a genuinely hard problem of its own making, and it is, to its credit, confronting that problem head-on rather than pretending it away. But "confronting it head-on" means selling the technology that undermines the pricing model on which its entire $40-billion empire was built, in a bet that the new model it is still inventing will generate more than the old model it is dismantling will lose. The Agentforce growth is real, the strategy is sound, and the eventual outcome may well vindicate the bet. What has changed — permanently — is that the question is now open. For twenty years, owning Salesforce meant owning the certainty of seat expansion. Today it means owning a wager on whether digital workers can be sold for more than the human seats they replace, refereed by a pricing model that has changed three times and a legacy base that is starting to shrink. The seat machine that built Salesforce is being dismantled by Salesforce, and what gets built in its place is the entire investment case — still under construction, still unpriced, and selling, every day, a little more of the AI that makes the old machine obsolete.

Salesforce got rich charging for every human who logged in, and it is now selling, as fast as it can, the digital workers built to make those humans unnecessary. It is the right move and a terrifying one at the same time, because the company's whole future now rests on a single unanswered question: whether an agent that does the work of three people can be sold for more than the three seats it just emptied — and nobody, least of all Salesforce's ever-changing price list, yet knows, which is exactly why the stock now trades like a question rather than the answer it was for twenty years.