The SaaS model was a brilliant invention. Pay monthly, access software from anywhere, no servers to manage. For twenty years it worked beautifully and minted some of the most valuable companies in history.
But SaaS was designed for a specific world: one where humans were doing the work, and software was the tool they used to do it. The pricing reflected that. You bought a seat because a person was sitting in it.
That world is ending.
When AI agents do the work — not assisting humans, but actually doing the tasks — the seat becomes a fiction. Nobody is sitting anywhere. The question is no longer how many people are using your software. The question is how much work got done.
That is a fundamentally different thing to charge for.
01THE SAAS TAX
Let us be honest about what the SaaS model became.
Somewhere between 2010 and 2020, “software as a service” quietly became “subscriptions as a default”. Vendors discovered that recurring revenue was extraordinarily easy to defend. Switching costs were high. Procurement cycles were long. Usage tracking was opaque. Nobody really knew how much of the platform they were actually using.
The result: companies are paying for vast amounts of software that nobody is using. Gartner estimates that organisations waste an average of 25% of their SaaS spend on unused or underused licences. Some analysts put the figure higher. Zylo, a SaaS management platform, found in its annual benchmark report that the average organisation uses just 56% of its licensed SaaS capacity.
That is not a rounding error. That is a structural tax on businesses, baked into the pricing model.
Vendr’s 2025 SaaS Trends report found that the average mid-market company now pays for over 130 SaaS tools. The average per-employee SaaS spend sits at around $18,000 annually and rising. For a company of 50 people, that is $900,000 a year in software subscriptions before you have written a single line of code or hired a single salesperson.
The SaaS vendors are not embarrassed by this. Bloat is the product. The more tools you are paying for, the harder it is to cancel any of them. Lock-in masquerading as convenience.
“SaaS was never about software. It was about locking customers into recurring payments and betting on their inertia.”
— Baz Furby
02AGENTS DON’T NEED A SEAT
Here is where the model breaks.
SaaS pricing assumes a human is logging in. It assumes that the “user” is a person who needs access to an interface, who benefits from training, who has a company email address and a calendar and a 1:1 with their manager. The seat exists because the seat is occupied.
But AI agents do not log in. They do not have seat counts. They do not need onboarding. An agent that runs a competitive intelligence workflow, enriches a lead, drafts an outreach sequence, and logs the result in your CRM does not consume a “user licence.” It consumes compute. It consumes API calls. It consumes tokens and tool uses and actions.
The unit of consumption has changed completely.
Andreessen Horowitz published research in 2024 noting that as AI agents begin handling tasks that previously required SaaS tools, the fundamental pricing question shifts from “how many seats?” to “how many actions?”. Sequoia Capital echoed this, predicting that outcome-based and usage-based pricing would become the default model for AI-native software within five years.
This is not a niche observation. It is a structural collapse of the assumptions that made SaaS pricing work.
The knock-on effect is severe for incumbent SaaS businesses. When companies can deploy agents to do the work that previously required five or ten licences, the revenue equation inverts. Usage goes up. Seat count goes down. The traditional growth lever — expanding seats as headcount grows — stops working the moment you replace headcount with AI.
“The question stops being how many people are using your software. It becomes how much work got done.”
— Baz Furby
03PAY FOR WHAT HAPPENS
If agents replace seats, the natural pricing model is credits.
Not credits as a cynical dark pattern — the kind where a gym sells you a membership and hopes you never show up. Credits as a genuine unit of work. You buy a block of compute-backed actions and you spend them on outcomes. When the credits are gone, real work happened.
This model aligns incentives in a way that SaaS fundamentally cannot.
Under SaaS, the vendor’s best-case scenario is that you pay, forget to cancel, and never actually use the product. Under an outcome-based credit model, the vendor’s revenue is directly tied to how much value they are delivering. Every credit spent represents a task completed, a lead enriched, a document generated, a report written. The vendor only gets paid when something actually happens.
That is a radically different relationship between price and value.
The implications for SaaS businesses are uncomfortable. If your growth model depends on seat expansion, you have a problem. If your retention model depends on switching costs rather than outcomes, you have a bigger problem. And if your margins are built on unused capacity rather than value delivered, you have the biggest problem of all.
The companies that survive this transition will be the ones that build pricing around what customers actually get, not around what they nominally have access to.
04WHAT COMES AFTER THE SEAT
The shift is already happening at the edges, and it is accelerating.
Intercom moved to outcome-based pricing for its AI agent, charging per resolution rather than per seat. Salesforce introduced usage-based tiers for Einstein AI. Stripe has always charged per transaction rather than per user. HubSpot, facing pressure from AI-native competitors, announced usage-based add-ons in 2025. Even Microsoft, the cathedral of seat-based licensing, is pricing Copilot on a usage tier rather than headcount alone.
These are not experiments. These are the first indicators of a structural repricing of the entire software layer.
The companies building natively on this model — charging for work done rather than access granted — have a structural advantage that incumbents cannot easily replicate. They can sell to companies that are contracting headcount while expanding output. They can grow revenue as AI adoption increases without needing to sell more seats into a shrinking workforce. Their interests are perfectly aligned with their customers’: when the customer gets more done, the vendor earns more.
“The businesses that survive the AI transition will be the ones that price on what you get, not what you pay for.”
— Baz Furby
05THE PLUGGIN MODEL
This is the thesis behind Pluggin.ai.
Pluggin is building what they describe as an agentic business operating system — a platform where AI agents handle real business workflows across sales, content, research, analytics, and operations. Ten agents. Seventeen integrations. Connected by shared memory, triggered by real events, supervised by you.
The pricing is credits. Not seats. Not per-user licences. Credits that represent actual work: a lead enriched, a report written, a competitor monitored, a customer onboarded. When you spend credits, something happened. When you run out, you know exactly what you got for your money.
This model strips out everything that SaaS pricing was designed to obscure. There is no unused-licence buffer. There is no per-seat expansion play based on hiring. There is no opaque feature matrix that exists to justify tier upgrades. Just: here is what an action costs, here is what it produces, here is how many you have left.
It is a more honest business model. And in a world where AI is doing the work, it is the only model that actually makes sense.
The wider implication is that if everything becomes a Pluggin — if the software layer is replaced by agentic action runners priced on outcomes — then the entire SaaS stack becomes redundant. Not immediately. Not without friction. But directionally, irreversibly.
Why pay $500 a month for a CRM that a human updates, when an agent enriches every contact, logs every interaction, and surfaces every warm lead automatically — and you only pay when it does? Why pay per seat for a content tool when an agent produces and schedules your content calendar, charged per post published? Why carry the SaaS tax when the outcome is available for a fraction of the price, delivered without anyone logging in?
The answer, increasingly, is: you will not.
06WHAT THIS MEANS FOR FOUNDERS
If you are building software today, the seat is a liability.
Not because SaaS companies are bad businesses — many will remain extremely profitable for years. But because the structural advantage of seat-based pricing is eroding fast, and the companies that do not voluntarily move toward outcome-based models will be forced to by competitors who already have.
Building on outcome-based pricing from day one changes everything about how you think about your product. You stop building features and start building workflows. You stop counting users and start measuring results. You stop defending your seat price against discounts and start competing on outcome ROI.
It is a harder thing to build. You have to actually deliver. There is no bloated margin from unused capacity to fall back on. But it is a more defensible business, a more aligned relationship with customers, and — as AI makes outcomes cheaper to produce — a far more scalable model.
The SaaS era was about selling access. What comes next is about selling results.
The businesses that understand this now will be the ones still standing in ten years.