The SaaSpocalypse: AI Agents began destroying the per-seat Subscription model's structural logic

The most successful business model in software history is being destroyed by its own customers. Not by a competitor with a better product. Not by an open-source alternative eating the margins. By the people who pay the invoices deciding, quietly and rationally, that they no longer need the thing they have been buying.

That thing is the seat.

The Seat Assumption

For two decades, the entire SaaS industry operated on a single foundational premise. I call it the seat assumption: the belief that software value is consumed by human users, and therefore the right way to charge for software is per human who uses it. More users, more value, more revenue. It was clean. It was predictable. It was the foundation of every SaaS financial model, every board deck, every investor thesis in enterprise software.

At Freshworks, I lived inside the seat assumption every day. Our product strategy was explicitly designed to increase seat counts within accounts. Features were evaluated, at least partly, on whether they would drive adoption by additional users in a customer's organisation. An integration that connected sales and support teams was not just a product improvement. It was a mechanism for turning a five-seat customer into a twelve-seat customer. The entire product roadmap had an embedded commercial logic: build things that create reasons for more people to log in.

And it worked. For years, it worked beautifully. Land a small team, prove value, expand to adjacent departments, grow the contract. The compounding maths of per-seat expansion was the engine that drove SaaS multiples to 18x, 19x revenue at the peak. Investors were not paying for today's seats. They were paying for the assumption that there would always be more seats tomorrow.

Nobody questioned the assumption. Why would you question a machine that prints money?

But that assumption contained its own destruction.

The Agent Substitution

When an AI agent does the work that a human with a software license used to do, the license becomes a cost without a user. This is not competitive displacement. Nobody switched to a better CRM. Nobody found a cheaper alternative. The work that required a person sitting in front of a screen simply stopped requiring a person.

Klarna announced it had replaced Salesforce's CRM with a homegrown AI system. The commentary focused on whether their internal tool was any good. But that is the wrong question. The right question is: what happens to a per-seat pricing model when the customer's strategy is to reduce the number of humans who need seats? But almost nobody in the SaaS industry was asking this question. They were still arguing about features.

I call this the agent substitution, and it is structurally different from any competitive threat the SaaS industry has faced. A competitor takes your customers. The agent substitution eliminates your unit of pricing. The customer stays. The seats disappear.

A founder I advise runs a B2B SaaS product that charges per seat. Solid product. Good retention. Growing steadily until three months ago, when one of their largest customers deployed AI agents across their operations team. The team went from twelve people to four. The subscription went from twelve seats to four. Overnight. Not because the product failed. Not because a competitor won the deal. Because the customer's AI strategy made eight of those humans, and their corresponding software licenses, unnecessary.

The product did not fail. The pricing model failed.

The Repricing

The market has noticed. The median EV/Revenue multiple for public SaaS companies has compressed from the pandemic peak of 18-19x to 5.1x. That is not a correction. It is a structural repricing of the entire application software layer. But the repricing is not uniform.

What the market is saying, in its blunt and unsubtle way, is that the seat assumption was not a business model. It was a temporary condition. A condition that held true only as long as software work required human operators. The moment that condition changed, the valuation framework built on top of it had to change with it.

But here is what most of the commentary misses. The agent substitution does not threaten all SaaS companies equally. It threatens the ones whose entire value proposition is wrapped around the interface layer: the CRM that is valuable because humans interact with it, the project management tool that organises human workflows, the communication platform that connects human team members. When AI agents can perform those interactions through APIs without ever touching a user interface, the seat is not reduced. It is made irrelevant.

The companies that are less exposed are the ones whose value lives in proprietary data, regulatory workflows, or domain-specific intelligence that an AI agent cannot replicate simply by accessing an API. But those are a minority.

Most of the SaaS market was built for human users doing human-scale work in human-operated teams. That world is not ending tomorrow. But the financial markets are pricing as though it is ending eventually, and they are adjusting now.

What the Numbers Cannot Show

At Freshworks, I remember sitting in product reviews where we would celebrate a customer expanding from eight seats to twenty. The room treated it like a product win. More users meant the product was working. More users meant more value delivered. But looking back, what we were actually celebrating was the seat assumption holding true for another quarter. We were celebrating a condition, not a capability.

The SaaS founder I advise is not panicking. She is rebuilding her pricing model around outcomes rather than access. But she told me something that stuck: "I spent three years optimising for seat expansion, and an AI agent undid it in a week." The speed is the part that catches people off guard. Competitive displacement happens over quarters. The agent substitution happens over a deployment cycle.

There is a strange irony in all of this. The companies that built the best products, the ones that made software so useful that organisations built entire workflows around them, are now the most exposed. Because those workflows are precisely the ones that AI agents can automate. The better your product was at enabling human work, the more clearly it demonstrated that the human was the variable, not the software. But nobody saw it that way at the time. Success has a way of hiding the assumptions it depends on.

Nobody planned this.

The per-seat model was not a mistake. It was the right model for a world where software was a tool that humans operated. But the world where software is operated by humans and the world where work is performed by agents turn out to be different worlds with different economics.

The question for every SaaS company is no longer how many seats they can sell. It is whether the seat is still the right unit of value at all. For a growing number of customers, the answer is already clear.

The quiet part is that most of us who built products inside the seat assumption never questioned it. It was the water we swam in. Some assumptions only become visible when they stop being true.