Agentic AI Is Killing Per-Seat SaaS Pricing

Here's the arithmetic quietly breaking the SaaS business model: if one AI agent does the work of five people, why would a company pay for five seats?

Per-seat pricing charges a company for each named user who can log into a piece of software, on the assumption that more users means more work getting done means more value worth paying for. It was always a proxy. You couldn't measure the value the software delivered, so you counted the humans and charged per head, and it held up because seats and value moved together. Agents snap that link. The work still happens; the logins don't. An agent has no named-user licence, doesn't show up in headcount, and doesn't politely buy another seat when the workload grows.

This isn't a forecast anymore. Deloitte's 2026 TMT Predictions put as many as 75% of companies investing in agentic AI by the end of this year. Gartner expects at least 40% of enterprise SaaS spend to move to usage, agent, or outcome-based pricing by 2030, with seat-based revenue sliding from 21% of the market to 15%, and reckons 40% of enterprise apps will ship a task-specific agent by the end of 2026, up from under 5% a year earlier. The agentic AI market itself goes from roughly $8.5 billion in 2026 to $35 billion by 2030 on Deloitte's numbers. Round the figures however you like; the direction isn't in dispute.

The companies in front of this have already moved, and they didn't agree on the answer. Intercom prices its Fin agent at $0.99 per resolution, so you pay when the customer's problem actually gets solved, not when a support rep logs in, and it rode that model from $1 million to over $100 million in ARR. Salesforce launched Agentforce at $2 per conversation, took the backlash for how unpredictable that felt, and now sells Flex Credits at about $0.10 per action instead. HubSpot's Breeze runs on credits too, roughly a dollar a conversation, with seats and usage folded into one hybrid bill. Three different answers to the same question: what do you charge for, once "a person" is no longer the unit of work?

Here's where most founders read a piece like this, nod, and file it under enterprise problem. That's the mistake. The seat was never only Salesforce's pricing mechanic; it's the default almost every founder reaches for when setting a price, because it's legible and it's what everyone else does. If your product is starting to do the work rather than just help a human do it, and most products are quietly heading there, then the thing you charge for is drifting away from the thing you deliver. That gap is the whole problem.

Pricing is one of the twenty-two commercial levers, and it's the one founders set once, early, off a competitor's page, and then leave alone for two years. It sits there looking finished while the product underneath it changes shape every quarter. The dangerous version of this isn't charging too little; it's charging for the wrong noun. Price per seat for software that's becoming agentic and you cap your own revenue at the number of humans in the room, right as the value you create stops depending on them.

So here's the exercise, and it takes about an hour. Open your pricing page and, next to each tier, write the actual unit of value the customer buys: not the feature list but the outcome, like a resolved ticket, a booked meeting, a reconciled invoice, or a qualified lead. Then ask whether your price tracks that unit or tracks the number of people who happen to log in. Where there's a gap, that's where your pricing is about to come under pressure, and where a usage or outcome line could sit alongside the seat rather than replacing it overnight. Most of the companies getting this right aren't burning the seat down; they run a base subscription with usage on top, so the floor stays predictable and the upside follows the work.

Outcome pricing isn't a free win, and anyone selling it as one is skipping the hard part. You have to define the outcome tightly enough that nobody argues about whether it happened (Intercom only bills a resolution the customer confirms, and backs it with a performance guarantee, precisely because the definition is where these deals live or die). Get the unit wrong and you've swapped a predictable bill your customer understood for an unpredictable one they resent. The work isn't picking usage over seats; it's naming what you're actually worth.

The founders who reprice early spend the next two years compounding on a model that grows with the value they deliver. The ones who wait get to explain to the board why revenue flatlined while usage tripled. The seat had a good run; it just stopped describing the thing you sell.