Per-seat pricing is dead. Most founders are mispricing the corpse.

Per-seat pricing made sense when software was a tool a person used. You paid per person, because value scaled with people. Agents break that cleanly: an agent does work without occupying a seat, so charging per seat means either you undercharge wildly (one seat, unlimited agent labor) or you're taxing a chair nobody sits in. The model is dying — seat-based pricing slid from about 21% to 15% of SaaS in a year while hybrid models jumped from 27% to 41%, and Gartner expects 40%+ of enterprise SaaS spend to shift to usage/agent/outcome pricing by 2030.

So far, so obvious. Here's where most founders go wrong.

"Charge for outcomes" is not a free lunch

The fashionable answer is outcome-based pricing: don't charge for access, charge per result. Intercom's Fin is the poster child — $0.99 per resolved ticket, scaling from $1M to $100M+ ARR. It sounds like the obviously correct future, and for the right product it is. But notice what outcome pricing quietly requires: you only get paid when the agent actually succeeds. You have just tied your revenue to your reliability.

That's a fantastic deal if your agent resolves tickets correctly 90% of the time. It's a slow-motion bankruptcy if it's right 60% of the time and you're eating the cost of every miss, every retry, every escalation to a human. Outcome pricing on an unreliable agent isn't a business model — it's you absorbing the compounding failure rate as negative margin.

Your pricing model is a bet on your own reliability. Outcome pricing says "I'm sure enough to put my margin on it." If you're not that sure, don't sign that bet.

The real principle: price follows who holds the risk

Forget the fad and think about reliability risk — the chance the agent is wrong — and who eats it:

• Charge per outcome when the agent is reliable enough that you'd happily carry the risk. You're saying "we're so sure this works that we only get paid when it does." That's the strongest possible trust signal — if you can back it.
• Charge per usage when the value is real but the outcome isn't guaranteed. The customer pays for the work done and holds the risk of judging quality. Honest, and it doesn't bankrupt you on misses.
• Charge hybrid — a platform floor plus usage or outcome on top — when you want predictable revenue without betting the company on a reliability number you can't yet defend. This is why hybrid is winning: it splits the risk.

The mistake isn't picking the wrong model. It's picking outcome pricing because it's trendy, on a product whose outcomes you can't yet stand behind — and discovering that you sold a guarantee you can't keep.

And don't just meter tokens and call it done

The lazy inverse is worth naming: bolting usage-based token metering onto old software and calling it "AI pricing." Passing your compute bill to the customer isn't a value model, it's a utility meter. Usage pricing should track value delivered, not your inference cost. If your price goes up because your prompts got sloppier, you've built a punishment, not a plan.

The bottom line

Seats are dying because agents don't sit in them, and everyone's racing to outcome pricing without noticing it's a leveraged bet on reliability you might lose. The right model isn't the trendy one — it's the one that puts the risk where it honestly belongs.

Price by who bears the reliability risk. Charge per outcome only when you'd stake your margin on the agent being right — because your pricing model tells the market exactly how much you trust your own product.