The ground under B2B software is moving faster than it has in 25 years.
And Wall Street has already started pricing it in...
Public SaaS growth has been decelerating for 16 straight quarters. In the most measurable AI workflows, agents are already resolving 70-85% of the work the seat was originally sold to authorize.
So the next question is no longer theoretical.
If the seat stops being the unit of value, what replaces it?
The smart money has converged on an answer. It’s worth understanding why, and why most of the companies trying to adopt it are about to fail.
The new contract
Outcome-based pricing is simple enough to explain in one sentence.
You pay when the software does the work (not when you log in).
A resolved support ticket. A closed lead. A processed invoice. A drafted contract. The unit of value isn't access, it's the outcome the software produced.
This is the model the seat was always pretending to be a proxy for.
The seat assumed a human would log in and do the work. Pay per seat, get the work. This held for 25 years because the assumption held.
The assumption no longer holds.
So the proxy gets priced directly.
Tomasz Tunguz put it this way:
"If an AI agent replaces an SDR who is compensated for meetings, then why not charge this way?"
Kyle Poyar said:
"The general trend is moving away from charging for access to software (seats) and toward a model of charging for the work delivered by a combination of software and AI agents."
The smart money is already there
Look at the pricing pages…
HubSpot moved its Breeze Customer Agent to $0.50 per resolved conversation in April. $1 per qualified lead on the marketing side.
Intercom's Fin charges $0.99 per successful resolution. If the customer escalates to a human or the AI fails to close the conversation, Intercom doesn't bill. The unit of value is the outcome, not the attempt.
Anthropic restructured Claude Enterprise this year from a $40-200/seat flat fee to $20/seat plus a mandatory usage commitment. The seat became the floor, not the ceiling.
ServiceNow disclosed publicly that 50% of new business revenue is now non-seat consumption.
Salesforce went "headless" (making its MCP layer available without a Salesforce seat at all).
SAP announced a wholesale shift away from SaaS subscriptions toward AI consumption pricing.
Adobe announced outcome-based pricing on its CX Enterprise product.
Outside enterprise:
Clay introduced dual-track platform plus token pricing.
Canva launched an AI Pass add-on with up to 40x more usage.
Every smart pricing page in software is converging on the same idea. Charge for the work. Stop charging for the chair.
Why the bolt-on doesn't work
Every legacy SaaS company is now bolting outcome pricing on top of seat pricing. The pricing page reads: a per-seat fee, plus a per-AI-resolution fee, plus an AI Copilot add-on per seat.
Three line items + three contracts and metrics that don't reconcile.
The customer pays for the seat and the resolution.
The bolt-on is a tax (and a scam).
Salesforce Agentforce is the cleanest case study. Three different pricing models in 18 months: $2 per conversation, then $0.10 per Flex Credit, then a hybrid bundle. The wobble itself is the story. They're trying to find one that doesn't cannibalize per-seat revenue.
The incumbents have a 25-year-old P&L built on seats. Every dollar of outcome revenue that isn't additive is a dollar that comes out of the seat line. The CFO won't approve a model that does that.
So the model gets layered on instead of replacing.
The result is outcome pricing on top of seat pricing. The customer still pays for the chair. They just also pay for the work the chair was sold to do.
Bolt-on outcome pricing isn't outcome pricing. It's seat pricing with a usage meter glued to the side.
The architectural question
If outcome pricing is the answer, the architecture matters more than the pricing page.
A real outcome-priced product has to be built for it from the foundation. The platform has to be free (because charging for the platform is just charging for the seat under a different name). The intelligence has to be metered natively because every resolution, draft, and closed lead is the actual unit of value.
And the AI has to be the product, not a feature glued onto a 12-year-old ticketing system.
You can't get there by re-pricing what you already built. You have to build the thing that gets priced this way from day one.
Twelve years of building, and we think we have it.
May 12
On Tuesday, May 12, we're showing up at SaaStr Annual with a manifesto and a protest sign.
We're going to do something Marc Benioff did 25 years ago.
You'll see why.
Cheers,