Operational Arbitrage, the Stewardship Model, and MTTI

The cost gap between what human labour costs and what an agentic stack costs to perform the same work is not a temporary market condition — it is a structural fact that compounds annually as LLM inference costs fall and human labour costs rise. Operational Arbitrage names this gap precisely. The Stewardship Model defines the architecture that captures it without eliminating human accountability: one operator governs an agentic stack as architect and exception handler, not executor. MTTI — Mean Time to Intervention — is the single metric that proves whether the capture is complete. When the core revenue loop runs for more than 72 hours without a human decision, the architecture is holding. When it does not, the dependency that is preventing full capture is still present — and measurable.

When an operator sees the cost differential between human labour and an agentic stack but has no precise vocabulary for it, the strategic response defaults to the familiar: deploy agents where the workflow is obvious, reduce headcount where tolerable, and treat the efficiency gain as the objective. What goes unmeasured is the gap itself — not the fraction recovered from it, but the fraction left in it. Operational Arbitrage names the full delta: not a point-in-time cost comparison but a structurally widening spread between two cost trajectories moving in opposite directions. A business that deploys agents into existing workflows without redesigning the Revenue Loop they operate inside captures some of the delta. The architecture that produced the original cost structure does not change. The gap narrows at one point and persists everywhere else.

The internal signature of a business capturing the arbitrage partially is a rising agent-to-task ratio alongside a coordination structure that is not shrinking. Output per workflow improves. The Coordination Tax — the budget consumed by meetings, handoffs, and status updates that keep human workers aligned — remains intact, because the agents have been inserted into workflows designed for human execution rather than rebuilt without that dependency at the centre. The Breakable Market test identifies the structural opportunity from the outside: a market where human labour represents more than sixty percent of gross margin is open to autonomous reconstruction at full capture. Inside a business operating in that market, the Workforce Arbitrage data puts a number on what is being left uncaptured. At Tier 1 tasks — routine, scripted, high-volume work — a single agentic workflow produces the equivalent of 37 to 50 human operators' throughput. The difference between that figure and what a partial deployment recovers is the Operational Drag the unreformed coordination structure sustains.

The architectural move most operators avoid is the one that changes the cost structure categorically rather than marginally: repositioning the human role from executor to governor. The Stewardship Model is not a cost-reduction strategy labelled differently. It is a commitment to a different dependency structure — one in which a single competent operator governs an agentic stack rather than coordinating a human team. The model makes Labor-to-Compute Substitution operationally credible: the steward is not displaced by the agents but defined in relation to them, handling the exceptions the logic cannot process and improving the architecture over time. What the steward governs compounds in value. What the agents execute costs less every year. Memo #03 argues the structural case precisely: the overhead a business carries is not inherited from its market. It is chosen in the architecture.

The Stewardship Model without a measurement of whether it is functioning as designed is an operating claim without verification. MTTI — Mean Time to Intervention — is that verification. It measures how long the core revenue loop runs before the steward is required to make a decision. Arco's target of greater than 72 hours is not arbitrary: at that threshold, the steward is intervening at most once every three days, and the system is governing itself across a full operational cycle. If MTTI falls below 72 hours, the steward is not governing — the steward is operating. The dependency that prevents Architectural Certainty is still present. Improving MTTI requires improving the architecture — not adding operators. Adding staff to a low-MTTI system raises cost without addressing the dependency. The dependency is structural and requires an architectural response. MTTI also resolves the Key-Man Risk concern that surfaces in acquisition and investment contexts: a business with a documented MTTI greater than 72 hours is not dependent on a specific individual to function. The system is the asset. The steward is replaceable. Memo #10 develops the full operational implications of the steward role across the business lifecycle — what it means to govern rather than operate, and what changes when MTTI is the metric the steward is accountable to.

A business that has captured the full arbitrage — through the Stewardship Model, verified by MTTI exceeding 72 hours — has a structural cost advantage that compounds without active investment. LLM inference costs continue falling. Human labour costs continue rising. The arbitrage widens on both sides without the business doing anything. The Human-to-Logic Ratio reaches the level that defines an autonomous business: a vanishingly small proportion of the operation still requires human execution. The 10:1 revenue-to-headcount advantage Arco holds as a confirmed portfolio target is the financial expression of that ratio at scale. Memo #05 develops the market selection framework in full — the conditions that make Operational Arbitrage wide enough to build a structural moat and defensible enough to hold it. A market that qualifies on both dimensions does not just offer an opportunity. It offers a compounding one.

Technology changes what is possible. Architecture determines who captures the gap — and whether the capture compounds.