I've spent my career selling software, which means I've spent my career selling seats. So believe me that I say this with affection: the seat is dying, and AI is what's killing it.
This isn't an analyst take. A few months ago we killed our own subscription model mid-flight and re-priced our live product around outcomes. Customers were on plans; money was flowing; we shut it off anyway. This essay is why — and what it taught us about how all software will be priced once it starts doing the work itself.
01What a seat actually measured
Per-seat pricing was never arbitrary. It was a brilliantly simple proxy chain: more people using the tool → more work flowing through it → more value delivered. Charge per person and you tax the value without having to measure it. The seat also matched the cost side: every user consumed support, storage, training. One metric, aligned both ways. CFOs could budget it; sales reps like me could forecast it. The entire SaaS industrial complex — comp plans, land-and-expand, NRR targets — is built on that proxy.
Then software started doing the work, and the proxy chain snapped.
Autonomous software inverts the correlation that made seats work: the better the product, the fewer humans need to touch it. Our best-case scenario is a customer who logs in once a week to glance at the dashboard and approve a budget move, while the software researches markets, sources leads, writes outreach, and fills the pipeline around the clock. Under seat pricing, our ideal outcome bills as one barely-active user. The proxy doesn't just drift — it points backwards. We'd be financially punished for building the thing we promised.
And the industry sees it. Bloomberg Intelligence projects subscription pricing falling from roughly 60% of software pricing models toward 30% over the next decade, with outcome-based pricing rising from 10% toward 60%. Every SaaS vendor adding "agents" is quietly discovering the same contradiction: they're selling labor-replacement priced by the laborer.
02We re-priced our own product. Here's the story.
Otto Outbound launched with the standard playbook — I built it, I believed in it: monthly tiers, volume-based plans, boosters, add-on packs. Eleven tiers, beautifully laddered. It worked, in the sense that people paid.
But every sales conversation had the same undertow. Prospects weren't asking which plan fits my team size — there was no team; that was the point. They asked one question, in different words: "what do I get for the money?" And our honest answer was a paragraph about tiers, quotas, slices, and boosters. When your pricing needs a paragraph, your pricing is wrong.
So we deleted it. All of it — tiers, per-play slices, daily caps, boosters, packs, trials. What replaced it fits in a sentence:
One credit = one contact, found, written to, and emailed for you. Buy credits when you want them. They never expire.
Prepaid wallet, volume discounts, optional auto-top-up. That's the whole model. No seats — add your whole team, we don't care, humans glancing at dashboards cost us nothing and you shouldn't pay for supervision. No monthly meter ticking whether the software worked or not. The unit you buy is the unit of work the software performs.
What happened next is the part I'd tell any founder: sales conversations got shorter. "€0.25 a contact, volume discounts, no expiry" answers itself. Risk perception collapsed — a prepaid wallet you can stop topping up any time reads as fair in a way a subscription with a cancellation flow never will. And one thing I didn't fully anticipate: outcome pricing forces honesty on the vendor. Under subscriptions, a mediocre month still bills. Under per-outcome pricing, our revenue moves with delivered work — every incentive we have points at the software performing. Your gym profits when you don't show up; your trainer doesn't. We became the trainer.
03"So usage-based pricing, you mean?"
No — and the distinction matters. The first wave of post-seat pricing was usage-based: per API call, per token, per compute-hour. That's pricing the inputs, and it quietly transfers all the risk to the buyer: an inefficient agent that burns 10x the tokens to do the same job bills 10x more. Buyers have noticed; CFOs are already writing horror stories about agent fleets tripling costs while revenue stayed flat.
Outcome pricing prices the output: the emailed contact, the booked meeting, the resolved ticket. Inefficiency becomes the vendor's problem — as it should be, since the vendor controls the machine. The ladder, in buyer-risk order: seats (pay for people) → usage (pay for the machine's effort) → outcomes (pay for the work) → and eventually, for some categories, results (pay for the business impact). Most of the market will settle on outcomes; results-pricing demands attribution most products can't honestly claim. We price the emailed contact, not the closed deal — Otto fills your pipeline, but your product and your closing still belong to you. Don't let a vendor charge you for work that's actually yours.
04How to buy software in the autonomous era
Flip this essay around and it becomes a buyer's checklist. When a vendor shows you their "agentic" product, ask:
- "What unit am I paying for?" If the answer is a seat, you're subsidizing their old model while they figure out their new one.
- "Does your price go up when your product gets better?" Outcome pricing scales with delivered work. Seat pricing scales with your headcount — which their product claims to reduce. Make them say it out loud.
- "What happens when it does nothing?" A subscription bills anyway. A wallet doesn't drain. This question separates vendors who believe in their autonomy from vendors who renamed their copilot.
- "Who pays for the model's inefficiency?" Token-metered pricing means you do. Per-outcome means they do.
Any vendor whose product truly does the work will love these questions. The ones who flinch are selling seats with extra steps.
05The seat was never the point
The seat was a proxy for work. For thirty years it was a good one. But proxies die when the thing they proxy changes shape, and work is changing shape faster than at any point in my career: from humans operating software to software operating itself, with humans steering.
Pricing always follows reality — usually a few years late, with the vendors who get there first taking the spoils. We chose to be early on our own product. It cost us some tidy recurring revenue and bought us something better: a price that tells the truth.
You can check the price yourself, and the product behind it, without talking to me or creating an account — which is itself a statement about how software should be sold now. Drop your URL at outbound.ottosoftwares.com. The first work is free. After that, you pay for exactly what it does.
— Thomas