Most business software is sold like a gym membership: priced per person per month. While buyers might sometimes wonder if they're getting the full value of their subscription each month, the business is delighted at the predictable, high-margin revenue stream that results. This model for delivering software has come to dominate since Salesforce pioneered it a quarter century ago, because it's simple, and it works. Each human who needs access to the system gets a seat. As organizations grow, they add humans and seats in lockstep, and the buyers and sellers all know what to expect.

Selling software on a subscription basis rests on three assumptions: work is performed by humans; load on the service is limited by the time and attention of those humans; and the business' margins are juicy enough to smear over any exceptions to the first two assumptions. The resulting businesses are among the most predictable, profitable, and easy to scale in history.

However, the agentquake roiling the industry is rapidly puncturing these assumptions. If, as we believe, there will soon be many more agents than humans doing knowledge work, charging per-seat is not going to work.

The Unreasonable Effectiveness of per-seat pricing

Businesses can afford to offer software on a subscription basis because human biology limits the load a single user can impose on the system. You can only click on so many CRM dashboards, email messages, or reacjis per hour. We click on only one UI element at a time, with our one (max two) mouse hands, during waking hours, across a small number of systems. Even if our usage is uneven, it is bounded; maybe someone out there uses Slack ten times harder than you, but nobody uses it one billion times harder. This simplifies everything! Cost of service per seat stays within a narrow range. System load is predictable. Work scales with headcount, so our per-seat price determines our margins. Et voila✨: we arrive at Enterprise SaaStopia.

Would you sell a gym membership to a robot?

In a system with a mix of humans and agents, you are no longer supporting a bounded, intermittent pattern of activity. As of this writing, agents lack labor rights. They do not sleep, eat lunch, or click a physical mouse and wait for feedback. They can also fork themselves into fractal trees that combinatorially explode the amount of work to perform. The peak-to-trough difference in load generated by different agents could easily be many billions, or greater, such that trying to express the price of the service as a function of the number of API keys you've issued is a fool's errand. At this point a "seat" ceases to be a fruitful unit of economic exchange.

Eventually providers of seat-based services will be selling dollars for 15 cents. Where will those nice, juicy margins come from?!

When SaaS personalizes to an organization

"I'll have my agents selfie your agents."

In the previous iteration of SaaS, organizations built workflows and processes around their software. In a world where agents can reshape workflows, create new connections between systems, or even modify the product itself, that relationship starts to invert. SaaS begins to personalize to the organization. You no longer need to design adaptive surfaces into the product. Agents—or fleets of agents—become that layer. What started as a fixed product begins to change shape. Instead of the organization adapting to the software, the software adapts to the organization. Not once, but continuously. Soon, you can imagine cross-SaaS agent coordination breaking down and automating more and more inefficient processes. Or humans with their own fleet of agents coordinating with a fleet of agents for a given SaaS product.

What comes next

Per-seat SaaS was built for a world where work was performed by humans, along product-defined workflows, in bounded increments, within fixed systems. As those conditions change, the foundation it rests on begins to shift. Software will need to anchor itself to something other than users—work, outcomes, or coordination between systems. We do not yet have a stable model for this. Isn't this just the most exciting?