The Cannibalization Trap

Institutions measure what their business model rewards. Those measurements become proxies, and over time, the proxy becomes mistaken for the outcome itself.

Seat-based pricing is one example. It made sense when human headcount was the unit of work. If more people were doing more work, more seats meant more value. That correlation is now breaking.

AI separates work from headcount — Reasoning from labor, and Value from seats.

The Cannibalization Trap is what happens when improving the customer's outcome makes the old business model smaller.

SaaS for Compliance

In the SaaS era, progress was measured largely as a function of scale. As we explored in From Builders to Orchestrators, engineering capacity was scarce and coordination was expensive.

When a VP of Sales buys 100 seats of a CRM, they aren't paying for 100 people to sell better. They are paying for a standardized process — to measure an account going south, to create a digital trail for the post-mortem, and to get dashboards that let leadership monitor what is happening.

In other words, the product supports the organizational structure that facilitates compliance.

Look at the titans of the SaaS era — Salesforce, Workday, ServiceNow. These platforms function primarily as Systems of Record. They create organizational hygiene by making the work visible, auditable, and measurable.

While that was valuable, it was never the same thing as creating the outcome. But it created a very clean unit of measurement: the people participating in the process. And that became the unit SaaS vendors learned to monetize.

The Digital Landlord

The model worked beautifully for vendors because it aligned revenue with the customer's inefficiency. If a customer's process was bloated and required 50 people to manage a workflow, the vendor got paid for 50 seats.

There was little incentive to automate the work. Automating the human out of the loop meant automating the revenue off the balance sheet.

SaaS companies became Digital Landlords. They rented you the desks — seats — but didn't care if work got done.

The Decoupling of Reasoning

In the Agentic era, AI agents move from recording work to participating in it. The System of Record becomes a System of Action - where the agent does not just record what happened, but reasons over the context, proposes what should happen next, and increasingly takes the action itself.

Reasoning and action are now decoupled from headcount, breaking the old proxy.

Consider a law firm using an AI agent to draft contracts. The AI allows one senior lawyer to do the work of five junior associates. If the vendor charges by the seat, something strange has happened. The vendor successfully made its customer more efficient — and in doing so, it helped the customer remove four users.

Building the Walled Gardens

This creates a predictable response. Legacy vendors cannot easily cannibalize their own business model. When AI agents begin performing work that was previously performed through seats, the vendor has an incentive to protect the unit it still knows how to monetize.

The response is predictable: vendors begin closing the system around themselves. External agents are blocked from accessing data. Proprietary agents become bundled into existing contracts. AI is forced through seat-heavy interfaces.

This is not necessarily irrational. It is rational behavior under a collapsing metric. The vendor is trying to keep the Input Metric alive in an Outcome World.

It is protecting the unit of value it knows how to monetize, even as the underlying unit of value is changing.

These strategies can work for a while. But they are all variations of the same move: defend yesterday's metric after the unit of value has changed. The market eventually moves on, whether the metric does or not.

The Definition Problem

So why haven't we switched to outcome-based pricing already? Because outcome-based pricing forces a conversation that most enterprises have avoided: what exactly is the outcome?

Seat-based pricing allowed organizations to be vague. "We bought the tool." "Adoption is high." "The CS team is keeping us warm." Those are safe statements. They don't require anyone to define whether the problem was actually solved.

Outcome pricing removes that safety. It is less like a Gym Membership — paying for access regardless of results — and more like Medical Treatment. You monitor the vitals. Did the problem get solved? Is the patient getting better?

The difference is accountability. With a gym membership, the transaction ends with access. But with medical treatment, someone is accountable for the result.

Outcome pricing scares the buyer because they have to define what success looks like. It also scares the seller because the seller has to guarantee something now. That is precisely why the transition is difficult.

The Challenge of Attribution

That does not mean we jump straight to revenue sharing. There is a natural progression.

First, measure what can be verified. Then, measure what can be attributed.

Level 1 — Verifiable Cost Savings. This is already happening. "Our Agent deflected 4,000 calls. At $5 per call, that is $20,000 saved." Easy to measure. Easy to verify. And it fits into an existing budget line.

Level 2 — Outcome Attribution. The harder question is what happens when the AI is not simply reducing cost, but influencing the outcome itself. Did the agent increase conversion? Did it improve collections? Did it shorten the sales cycle? Did it prevent churn?

And more importantly: how much of that outcome can be attributed to the agent? This is where the economics change.

Now the conversation is different. The vendor is no longer selling access to software — It is participating in the outcome.

The ultimate litmus test for an AI vendor becomes: "If your AI makes me twice as efficient, does your revenue go up or down?" If the answer is "down," the business model is fighting the product.

Proxy Collapse

Every institution eventually builds proxies for value: Seats, Hours, Utilization, Adoption, Compliance.These were reasonable measures when value was created through human effort. They worked because the metric correlated with the thing the institution actually cared about.

That relationship is now breaking as AI separates work from headcount — Reasoning from labor, and Value from seats.

Agentic Economics

The SaaS era was built on Distribution Moats and Seat-Based Rents. The Agentic era will be built on Attribution Moats and Outcome-Based Value.

For the buyer, the question becomes simple: what did the system actually change?

For the vendor, the question becomes harder: can you price yourself on the value you create without destroying your own economics?

Legacy vendors will have to cannibalize their own business models or be displaced by shifting market forces. This is not necessarily a failure of the SaaS model. It is what happens when the unit of value changes.

The SaaS era monetized the Tool. The Agentic era can monetize the Work. But only if institutions have the Trust Budget to let the system perform it.

Agentic AI demands Agentic Economics. If the software does the work, the economics must eventually follow the work.

That is the deeper transition. The technology creates the possibility of a better outcome. The inherited metric can make that outcome economically impossible.