Who Gets the AI Dividend?
For most of the software industry's history, pricing has been attached to something relatively easy to count. Users and seats are familiar examples, but vendors also charge by devices, locations, transactions, data, API calls and consumption. Whatever the unit, it eventually becomes arithmetic in somebody's financial model: volume × price drives the vendor's revenue forecast, while that same equation lands as an expense in the customer's budget. The models vary, but they share an important assumption—that the thing being counted reasonably tracks how the customer uses and values the product.
AI is beginning to complicate that relationship. If an AI agent can perform work that once required ten employees, a customer may need fewer seats while getting substantially more value from the software. But the same disruption can occur with other pricing units as automation changes transaction volumes, API consumption, compute requirements or even the number of applications a customer needs. Suddenly, the question isn't simply whether seat-based pricing survives. It's whether the thing we're counting still corresponds to the value being created.
When Success Breaks the Pricing Model
Consider a vendor earning $1 million annually from 500 users at a customer. Suppose AI enables 100 employees, assisted by agents, to accomplish roughly the same work. If pricing remains tied entirely to seats, the customer may capture enormous productivity gains while the vendor watches revenue decline precisely because its technology worked. That's an uncomfortable business model.
Vendors aren't likely to ignore that for long. We're already seeing experimentation with usage, consumption, credits, outcomes and hybrid pricing because everyone is trying to answer a deceptively difficult question: if software is doing more of the work, what exactly are we selling? Access to the software is one possibility. The work performed is another. The economic outcome created for the customer may be another still.
Who Gets the Productivity Dividend?
This is where an interesting form of pricing arbitrage appears. Imagine an AI agent costs a vendor $2,000 to operate but enables a customer to avoid $80,000 of labor expense or produce the equivalent additional output. Somewhere between those numbers sits a considerable pool of economic value. The customer would understandably like to keep as much of that productivity dividend as possible, while the vendor can reasonably argue that its technology created it.
Neither side gets to decide the answer alone. A vendor might calculate that an agent creates $2 million in annual customer value and conclude that charging $500,000 is a bargain. Customers may agree at $250,000, hesitate at $500,000 and start evaluating substitutes at $750,000. Competitors will have opinions too. As similar capabilities become more widely available, today's scarcity premium may prove temporary.
This is why the AI pricing problem isn't solved simply by replacing per-seat pricing with "value-based pricing." The market still determines what it will bear. Willingness to pay, differentiation, switching costs, alternatives and competition determine how much of the economic surplus a vendor can actually capture. In other words, AI may be rewriting software pricing, but it hasn't repealed price elasticity.
The Forecast Has to Change Too
There's another problem hiding inside all of this: the forecast. A company can spend months explaining how AI will transform the way its customers work and then hand Finance a five-year revenue model that assumes absolutely nothing changes. If the forecast assumes today's pricing metrics will continue behaving exactly as they do now while the product strategy is explicitly designed to change how customers perform and consume the work, the demand curve may be resting on assumptions that are already becoming obsolete.
That doesn't necessarily mean the revenue opportunity is smaller. It means the economic model may need a different unit of value. A successful AI product might generate fewer seats but more consumption, higher-value transactions, outcome-based fees or entirely new revenue streams. The important part is recognizing those possibilities before the old assumptions quietly become the new forecast.
That uncertainty belongs in the financial model, not in the explanation after the numbers miss. Product leaders should be showing Finance how different pricing architectures affect revenue, adoption, cost-to-serve and margins. A seat-based scenario might sit alongside consumption and hybrid models, with different assumptions about adoption and willingness to pay. The objective isn't to predict exactly which model wins. It's to show where the economics are sensitive and what evidence would tell you that the market is moving.
The Market Gets the Final Say
Ultimately, the right AI pricing model won't be discovered in a spreadsheet. It will be discovered in the market. Vendors will test different units of value, customers will reveal what they're willing to pay, competitors will apply pressure, and demand will respond as prices move.
AI may change who performs the work, how much that work costs and where the resulting economic value accrues. But the fundamental pricing question remains remarkably familiar: what will the market bear, and how does demand change when the price changes? The companies that answer that question with evidence rather than yesterday's assumptions will be better positioned to find the equilibrium that matters: value for the vendor, value for the business, and value for the customer.