Go-to-market is a craft.
Pricing is a discipline.
Enterprise deals are won by both.

Per-seat pricing was never really about seats. It was a proxy: a crude but reliable stand-in for value in a world where software made humans more productive, and more humans meant more value.

AI inverts the proxy. When an agent resolves the support ticket, drafts the contract, or reconciles the ledger, value delivered goes up while seat counts go down. A vendor pricing per seat is now paid less for building a better product. The best AI companies will cannibalize their own seat counts, and the pricing model must be rebuilt before that happens, not after the renewal where a customer asks why they're paying for 400 licenses and 60 humans.

The uncomfortable answer is that there is no single successor metric. Work units, resolved outcomes, consumption with committed floors, value-share on measurable savings: each fits a different buyer, budget process, and trust level. The craft is in matching metric to market. Enterprises will not sign an uncapped outcome-based deal from a Series B vendor, but they will sign a hybrid: platform fee for predictability, usage for scale, an outcome kicker where ROI is provable. The companies that get this right won't just protect revenue as seats disappear. They'll capture, for the first time, the actual economics of the work their software does.

Software spent two decades enjoying an economic miracle: 80%+ gross margins that nobody had to manage. AI ended the miracle. Marginal cost is back, and most pricing models haven't noticed.

Every AI feature ships with a meter running. Token consumption varies wildly by customer, use case, even prompt style. Charge a flat subscription against a variable cost and you have written your customers a free option. Your heaviest users, often your flagship logos, become your least profitable accounts. I've seen AI products where the top decile of users consumed 20x the median. Under flat pricing, that's not a power-user success story; it's a structural margin leak the CFO will eventually find.

This is a pricing-design problem, and it's solvable, but only deliberately. It means knowing cost-to-serve at the account level, not the blended average. It means fair-use thresholds, committed-usage tiers, and overage terms that are commercially graceful rather than punitive. It means deciding which margin risks you keep (model efficiency gains you can capture) and which you share (customer-driven volume). And it means resisting the venture-funded temptation to price AI at a loss, because negative-margin revenue doesn't scale into a business; it scales into a bigger problem. In the AI era, gross margin isn't an accounting outcome. It's an architecture choice.

Enterprise procurement has priced software for thirty years. It has priced intelligence for about three. The result is a negotiation table where neither side fully understands the cost structure being contracted, and the vendor usually pays for the ambiguity.

Large buyers are importing cloud-era tactics into AI deals: flat, all-you-can-eat pricing while their own AI usage grows unpredictably; multi-year price locks on services whose underlying costs (model pricing, context lengths, agentic workloads) move quarterly; benchmarking clauses that treat a reasoning system like a commodity SaaS module. Vendors, desperate for the logo, sign terms that transfer all of the usage risk and none of the upside. Three years in, the contract everyone celebrated is a margin trap nobody can exit.

The negotiation frontier is shifting from price level to risk allocation. Who absorbs model-cost movements, and who benefits when inference gets cheaper? What is a “unit” of AI work, contractually, and who measures it? Does a committed-spend structure flex across models and use cases, or lock the buyer into yesterday's architecture? These clauses, not the headline discount, determine deal economics. My advice to vendors: negotiate the meter before you negotiate the number. And to the PE firms underwriting these revenue streams: read the AI terms in the top ten contracts first. That's where the real quality of earnings now lives.