.png)
In brief: The predicted collapse of per-seat software pricing did happen in early 2026 (~$2 trillion in market value wiped out), but the winner wasn't pure outcome-based pricing as analysts forecast — it was the hybrid model: a fixed subscription plus a variable usage/outcome fee. Salesforce's Agentforce shows the playbook, running three pricing models at once and letting customers choose, while shifting the basis of value from access (seats) to actual work done.
Last summer, the Boston Consulting Group was still phrasing things in cautious future tense: AI agents would rewrite how B2B software companies make money, and pricing based on user seats — where a company pays for as many people as use the software — would be shaken. Anyone in the market knows this is no longer a forecast.
In the first months of 2026, roughly USD 2 trillion in market value vanished from the software sector, in a sell-off the press dubbed the "SaaSpocalypse." According to Bain & Company, the market froze, and AI is a structural threat to seat-based models in certain use cases. And there are already vendors reporting slowing seat sales as their customers become more efficient.
I don't want to write about the panic. Rather about what the market actually did after the initial shock — and why exactly that. Because the answer isn't what last year's predictions would have led you to expect.
The trigger wasn't a single event but a realization. The logic, as Bain summed it up, is brutally simple: if an agent does the work of ten people, then ten human seats are no longer needed — and the per-seat revenue model collapses.
And the numbers became concrete:
This is worth putting in perspective: Bain and others stress that this isn't a collapse but a repricing — a correction of investors' earlier, overly optimistic assumptions. Software is still a good business: customers are slow to switch, and margins are high. But growth is now shifting toward AI features and agents, and that requires new investment, which not every company can pull off.
Last year's consensus — including the BCG piece — held outcome-based pricing to be the end state: the buyer pays only for tangible business results, not for access. Perfect on paper. In practice, though, the data from the first half of 2026 shows something else.
The market chose the hybrid: a mix of a fixed subscription and a variable, usage- or outcome-based fee. According to a Pilot survey, within a single year only 15% of SaaS companies still priced purely on a seat basis, down from 21%. But it wasn't pure outcome-based pricing that took its place — it was this hybrid, now used by 41% of companies instead of 27%. The same source adds that companies using hybrid pricing report 38% higher revenue growth and 38% better net revenue retention (NRR) than purely subscription-based companies.
Why? Because pure outcome-based pricing ran into exactly the walls BCG predicted:
If there's one story worth knowing by heart on this, it's Agentforce's. Salesforce shipped three different pricing models for the same product in about 18 months. Not because it couldn't decide what it wanted, but because this turned out to be the smartest move in an immature market:
Salesforce didn't kill the old models: all three run simultaneously. It lets the customer choose how they want to pay.
And it all seems to be working. In the first quarter of fiscal year 2027, Agentforce ARR jumped to USD 1.2 billion, up 205% year over year: the fastest-scaling AI product line any enterprise software company put up in 2026. The architectural difference Benioff emphasizes is that Agentforce isn't priced per seat but on the basis of work units (Agentic Work Units) — that is, on the task actually performed, not on access.
For credibility's sake, the gloomier reading (the "bear case") belongs here too: Salesforce stock was the weakest component of the Dow in 2026, down roughly 32% since the start of the year. The market is still pricing in the possibility that seat erosion could move faster than Agentforce revenue can replace it.
One of BCG's best warnings was that cost-plus pricing — slapping a plain margin on top of token costs — makes for a revenue rollercoaster: as model costs fall, which is a baseline trend in AI, so does revenue, unless usage grows exponentially.
By 2026 this had grown into a real margin problem. Unlike traditional software, AI comes with significant variable costs: every agent task eats compute, API calls, and token processing. This squeezes the executive from two directions: the variable cost calls for protection (which pulls toward usage/outcome), but the unpredictability makes the customer cling to a fixed frame. Hence the dominance of the hybrid: a fixed base fee plus a variable consumption band, which became the prevailing model of the transition in 2025–2026.
The golden age of software sales was about how many people sit in front of the screen. In 2026 it turned out that this measured the wrong thing: access, not value.
The lesson, though, isn't last year's script — no single model swept the rest away. The ones doing well are those who run multiple models at once, let the customer choose, and shift the center of gravity gradually and predictably from access to actual work.
The financial figures are snapshots and do not constitute investment advice.