The Unit Economics of Autonomy

Philippe Van Caenegem identified something that most of the AI industry has not yet internalized. The economics of intelligence are not the economics of software. SaaS companies sell access. Agentic companies sell outcomes. And the unit economics of outcomes follow entirely different laws.

The concept Van Caenegem introduced — intelligence yield — is deceptively simple: the number of successful decisions produced per dollar of system cost. But its implications are profound, because it forces a company to account for something that traditional software economics ignores entirely: the cost of reasoning itself.

The hidden cost structure of thinking

In a SaaS model, the marginal cost of serving an additional user approaches zero. The code runs. The database queries execute. The infrastructure scales predictably. This is why SaaS margins are well-understood and why the industry converged on 20–40% operating margins as the standard for healthy businesses.

Agentic systems break this model completely. Every decision an agent makes costs money — in inference, in compute, in the human oversight required to validate high-stakes outputs. The marginal cost does not approach zero. It approaches whatever the cost of reasoning happens to be for that particular problem. And reasoning costs are wildly variable.

This is where Van Caenegem's framework becomes essential. He proposes modeling the unit economics of an agentic system through what he calls the Intelligence ROI formula — a calculation that places reasoning cost at the center of the margin equation, not as an afterthought but as the primary variable that determines whether a company is profitable or burning capital on computation that produces nothing.

Reasoning waste as the central risk

The most dangerous concept in Van Caenegem's framework is reasoning waste. In traditional software, waste is relatively benign — an unused feature costs storage but not ongoing computation. In agentic systems, waste is active. Every redundant token generated, every low-confidence reasoning branch explored, every speculative inference that leads nowhere consumes real resources and produces negative returns.

This is why reasoning waste is not merely an efficiency problem. It is a margin problem. A company whose agents reason wastefully will see its costs scale faster than its revenue — the precise inverse of the SaaS scaling dynamic that investors have spent two decades learning to love. The agentic company that cannot control reasoning waste does not have a technology problem. It has a business model that collapses under its own computational weight.

The firms that understand this are already pulling away. High-performing agentic companies are achieving operating margins of 40–60% — significantly higher than traditional SaaS — not because their technology is fundamentally cheaper, but because they have learned to stabilize reasoning costs. They know which problems are worth reasoning about deeply, which can be resolved with shallow inference, and which should not be attempted at all.

The Outcome Dependence Ratio

Van Caenegem introduces a second metric that completes the picture: the Outcome Dependence Ratio, or ODR — the percentage of a customer's operations that are executed entirely by AI agents. This is not a vanity metric. It is the single best predictor of whether an agentic company has achieved genuine product-market fit or is merely providing an expensive copilot.

A low ODR means the system assists. A high ODR means the system decides. The economic difference between these two states is not incremental — it is categorical. An assisting system competes with other tools on features and price. A deciding system competes on outcomes and becomes structurally embedded in the customer's operations. The switching costs become enormous not because of lock-in, but because the customer's own decision-making capacity has been rebuilt around the system's intelligence.

What this means for ogram

We recognized the implications of Van Caenegem's framework early because they describe exactly the economics we are building toward. ogram does not sell software access. We sell intelligence yield — successful strategic decisions per dollar of system cost. Every architectural choice we make is disciplined by this metric.

Our reasoning architecture is designed to eliminate waste at every layer. We do not run speculative inference chains hoping for insight. We define the decision space, constrain the reasoning to what is consequential, and produce calibrated outputs where every computational step can be justified against the value of the decision it informs. This is not an engineering preference. It is an economic requirement.

The consulting industry, by contrast, has no concept of reasoning waste because it has no concept of reasoning cost. When McKinsey deploys twelve analysts for eight weeks, nobody asks what percentage of their cognitive effort produced actual insight versus what percentage was spent reconstructing context, navigating politics, and formatting slides. The question has never been asked because the business model does not require it. The client pays for time, not for yield.

The shift from selling time to selling intelligence yield is not a business model innovation. It is an economic phase transition — and it will be as decisive for strategy consulting as quantitative trading was for equity research.

Van Caenegem's insight is that the companies that master intelligence yield will not just outperform SaaS companies on margins. They will create an entirely new category of economic entity — one where the product is not software, not advice, not access, but the measurable production of good decisions at scale. That is the company we are building.