Software Sold Seats. Cloud Sold Consumption. AI Will Sell Outcomes.

Enterprise software has repriced itself twice in my career. Perpetual licenses became seats when software moved to the cloud and value attached to the people using it. Seats became consumption when infrastructure took over and value attached to the compute you burned. Each shift felt radical, then obvious — because pricing was chasing the same thing both times: the unit of value.

The third repricing has started, and it will be bigger than either of the first two, for a simple reason. For the first time, the software is doing the work.

Seats break when agents do the work

A seat is a proxy. It says: value scales with the number of humans using the tool. That proxy held for thirty years, and agents break it in one move. When an agent resolves the tickets, reconciles the invoices, or runs the outreach, there is no seat — there is work, done.

Worse, per-seat pricing now punishes the vendor for being good. Build an agent effective enough to let a ten-person team do the work of thirty, and under seat pricing you just cut your own contract by two-thirds. Every software CFO can see the collision coming: the better the product, the smaller the renewal. A pricing model that taxes your own product improvements is not a model. It is a countdown.

Consumption prices activity, not value

The reflexive answer is to meter it — tokens, calls, agent-hours. Cloud trained everyone to think this way, and it is the wrong lesson here. Consumption prices activity, and with agents, activity and value can point in opposite directions. An agent that loops inefficiently through a task burns more metered units than one that solves it cleanly — so the meter literally monetizes inefficiency. The buyer carries all the risk of a chatty agent; the vendor gets paid more when the product works worse.

Buyers already have scar tissue. Every CFO who has opened a surprise cloud bill knows what an unbounded meter feels like, and they are not eager to sign another one — this time attached to software that spends autonomously. Consumption is a fine cost-recovery layer. It is not what the work is worth.

Outcomes are the honest unit

The honest unit is the finished thing: the ticket resolved within policy, the invoice processed and posted, the qualified meeting that actually happened, the claim adjudicated correctly. That is what the buyer wanted all along — seats and meters were only ever proxies for it, tolerated because nobody could measure the real thing at scale.

Agents change the economics of measuring. Work that flows through software is work that can be counted, evaluated, and priced at the unit where value actually lands. When the proxy and the real thing cost the same to measure, the proxy dies. It always has.

Make no mistake about what outcome pricing means for vendors: it is underwriting. You are no longer selling access to a tool; you are pricing the risk that the work gets done to spec. That is a harder business — and a far better one, because the vendor who can carry that risk profitably is the vendor who has actually proven the product.

Outcome pricing demands outcome accountability

Here is the part most pricing decks skip. You cannot invoice an outcome you cannot prove. The whole model stands on three requirements, and they are contractual and architectural before they are commercial:

Definition. What counts as the outcome — written down, edge cases included. “Resolved ticket” means resolved to whose standard, within what window, with what re-open rate? The outcome definition becomes the new battleground of the MSA, the way uptime SLAs once were.

Attribution. Did the agent produce the outcome, or ride along while a human did? Agent-produced and agent-assisted are different products at different prices, and a vendor who blurs them will lose the account the first time a buyer audits the split.

Evidence. A provable record that the outcome occurred — and that what the agent did to produce it matched what it was authorized to do. This is where the money model meets the trust model from the first essay in this series: the same decision evidence that satisfies your auditor is what lets a vendor invoice with a straight face. I’ve been working on the evidence half in the open — a proposed OCSF extension for recording authorization decisions at the data element is on GitHub.

What it does to go-to-market

I have spent twenty-five years building revenue engines, and outcome pricing rewires nearly every part of one. The demo stops mattering; the measurement period replaces the pilot, because the pilot’s job is now to calibrate the price. Sales becomes an underwriting exercise: qualification means understanding the buyer’s baseline well enough to price the delta. Pricing teams turn actuarial. The CFO joins the buying committee on day one, not at redlines — because the contract is now a measurement spec with a payment schedule attached.

And the champion’s job changes in a way vendors should love: instead of defending a license fee at every budget review, they walk into the QBR with a ledger of outcomes both sides already agreed were real. The renewal conversation becomes arithmetic.

Adoption metrics invert along with the pricing. Seats and daily logins were always proxies for humans at work; once agents do the work, the meaningful unit of adoption is delegation depth — how many governed, completed outcomes the enterprise entrusts to the system, at what value, cost, and reliability, under what authority, with what evidence. A million users is a vanity metric. A million outcomes you can defend is a business.

What to do now

Five moves, whichever side of the table you sit on:

Instrument before you price. You cannot sell outcomes you are not measuring today. Vendors: build the outcome ledger into the product now, while you still sell seats.

Define the outcome contractually. Write the definition, the window, the quality bar, and the dispute path. If the definition takes more than a page, the product is not ready for the model.

Keep attribution honest. Separate agent-produced from agent-assisted in the data from day one. The split will be audited; volunteer it first.

Share the ledger. One record of outcomes both sides can read beats two dashboards that disagree in the renewal meeting.

Run one outcome-priced deal in 2026. Small scope, real money, real definitions. Buyers: even on seat or consumption paper, demand outcome definitions and evidence clauses now — it forces the discipline before the pricing arrives.

Who wins the repricing

The vendors who win will not be the ones with the best model. Models are converging into table stakes exactly the way intelligence is. The winners will be the ones who can prove the outcome — because proof is what lets you price it, and pricing the outcome is what lets you capture the value the agent actually created.

Seats measured people. Consumption measured activity. Outcomes measure the truth — and the truth needs evidence.

This is the second essay in Field Notes on the Agentic Enterprise. Essay 01: The Agentic Economy Runs on Authority. Next: “Verification Is the New Bottleneck in Enterprise AI.”