The Operator Log, Episode Three. How We Think. Overhead Is a Design Choice. Operational overhead is not the cost of growth. It is the cost of building wrong.
Two episodes in, two architectural claims. First: an autonomous business decouples revenue from headcount by design, not by accident. Second: agentic architecture is what makes that decoupling structurally possible. Today we deal with what you are paying every quarter because you have not built that architecture yet. Not overhead as an operational burden. Overhead as a structural indictment. Most companies treat rising overhead as evidence their business is scaling. Arco treats it as evidence their architecture is failing. Those two interpretations produce different companies — and the difference compounds every year the wrong interpretation holds. This is The Operator Log.
The default response to growth in a legacy firm is hiring. Revenue goes up. Complexity goes up. Headcount goes up. This sequence feels like logic because in a business designed around human execution, it is logic. More complexity genuinely does require more humans to manage it — when the architecture depends on humans to manage it. The assumption is never questioned because it has always been true within the constraints of the model that produced it. The error is not the conclusion. The error is treating a constraint of one architectural model as a law of nature.
Here is what that cycle actually produces. Every new hire adds a coordination surface. Someone needs to brief them on the context they were not present for. Someone needs to align with them on decisions that were already made before they arrived. Someone needs to review their work before it moves forward. Someone needs to wait on them when the next step depends on their output. Each of those interactions is a cost. Not a cost that appears on any individual line item — a cost that spreads invisibly across every workflow the new person touches. The Coordination Tax compounds. It does not add. It multiplies.
And here is the trap's deeper mechanism. Each shortcut taken to move faster — each process left undocumented, each approval gate left in place because removing it would take a week the team does not have, each piece of logic left in a human's head instead of encoded in the system — becomes structural overhead that accumulates with scale. When the business is small, these shortcuts are invisible. They are absorbed by the people who know where the bodies are buried. When the business grows, they surface as debt. The cost of re-architecting a system built under those conditions — of going back and replacing human-dependent logic with deterministic architecture at scale — is what we call the Rebuild Tax. It is not a one-time cost. It is the compounding interest on every architectural decision that prioritised speed over design integrity. Episode four is built around it.
Research from McKinsey's twenty twenty-five State of AI report shows that high-performing organisations distinguish themselves by redesigning workflows rather than accelerating them — reclaiming fifteen to twenty hours per week per process by eliminating the process entirely, not by making it faster. Most organisations are doing the opposite. They are using AI to run the existing architecture at higher speed. Which means the Coordination Tax is still accumulating. It is just accumulating faster.
The standard counter-argument to all of this is that complexity genuinely requires human judgment. That edge cases, relationship decisions, and strategic calls cannot be handled by a system. This is a reasonable position. It is also, as the live memo for this episode states, a description of a different problem: the failure to design the system well enough. Not every decision belongs in the autonomous tier. We established the T-Tier Hierarchy in Episode one precisely because some decisions remain human-centric by design. The question is not whether any human judgment is required — it is whether the humans in the system are doing judgment work or coordination work. Coordination work is not judgment. It is overhead. And overhead is eliminable. Incumbents solve complexity with more people. Arco solves complexity with better logic. Those are not two approaches to the same problem. They are two different theories of what a business is.
The Coordination Tax has a number. In a traditional service business, it consumes between twenty and thirty percent of the operating budget — not by producing anything, but by humans aligning with other humans. That range is validated by both McKinsey Global Institute analysis and Harvard Business Review research on operational costs. It is not a rounding error. It is a structural feature of every business designed around human handoffs.
To understand what that number actually represents, it helps to make it concrete. Consider a services firm that produces a weekly status report for each of its clients. In the legacy model, the account manager drafts the report. The project lead reviews it for accuracy. The operations head checks it for format compliance. A senior account director signs off before it goes to the client. That is four people, three review cycles, and somewhere between two and four days of calendar time — for a piece of work whose output is entirely deterministic. The data exists. The format is fixed. The logic for what goes in each section does not change week to week. The only reason four humans touch this document is that the system was never designed to produce it without them.
In an autonomous business, an agent pulls the relevant data from the live systems, applies the report template, formats it according to the client's specifications, and sends it. Zero handoffs. Zero review cycles. Zero calendar latency. The same deterministic output, produced without the coordination overhead that the legacy model requires structurally. That weekly report, multiplied across every client and every reporting cycle, is where the twenty to thirty percent goes. It is not going to strategy. It is not going to growth. It is going to the administrative layer that exists solely because the business was designed to need it.
This is not waste in the colloquial sense — something that could be cut with a memo. It is structurally necessary in a business built around human execution. You cannot remove the Coordination Tax without removing the human-in-the-loop dependencies that generate it. And most companies are not willing to do that, because removing those dependencies requires redesigning the processes that depend on them. So they pay the tax every quarter and call it the cost of running a business at scale. There is a distinction worth holding here, because it determines what is actually eliminable. Not all overhead is administrative. Every system requires maintenance — infrastructure, compute, stewardship, security. That is operational overhead. It is structural and intentional. What Arco targets is the administrative layer — the tier of human management that exists solely to coordinate other humans rather than to produce output. In an autonomous business, that layer is replaced entirely by agentic handoffs.
Agents share a single source of truth. They do not need to be briefed. They do not need to align. They do not wait on each other. The logic that coordinated the humans is encoded in the architecture, and the coordination overhead goes with it. The Coordination Tax does not get optimised in an autonomous business. It gets structurally eliminated. That distinction — between reducing a tax and removing the structure that generates it — is the difference between an AI transformation programme and a reconstruction.
At Arco, we do not measure productivity. Productivity is the right metric when human output is the ceiling. In an autonomous business, human output is not the ceiling — architectural design is. The metric that replaces productivity is Operational Drag. The precise definition: Operational Drag is the ratio of non-revenue-generating tasks to total compute within a business system. It measures how much of the system's total operational capacity is consumed by work that does not directly produce output. In a legacy firm, this ratio is structurally high and largely invisible — masked by the appearance of activity. Everyone is busy. The Coordination Tax is running at twenty to thirty percent. The business looks like it is operating at full capacity, because it is. The problem is that a significant fraction of that capacity is consumed by the architecture managing itself rather than the architecture serving customers.
The architectural mandate for every Arco business is to hold Operational Drag below five percent of the total operational cycle. That number is not a performance target to optimise toward after the business is running. It is a design constraint that shapes how the business is built before it launches. If a proposed workflow requires constant human intervention to maintain its current scale, the architecture does not meet the mandate. The workflow is redesigned until it does, or it is classified as Tier three — one of the decisions that remains human-centric by design — and handled accordingly.
To make the five percent mandate concrete: consider what the human role in an Arco business actually looks like when the architecture is holding. In a given week, the person overseeing the operation is not approving reports. They are not reviewing agent outputs before they ship. They are not sitting in alignment meetings. They are auditing system health — checking whether MTTI is holding above seventy-two hours, reviewing the exceptions the agents surfaced and could not resolve autonomously, updating the logic rules for any exception that recurred more than once, and expanding the scope of agent authority in the tiers where the architecture has proven stable. That is a week of work. It is not a week of management in the legacy sense. It is a week of stewardship.
That role — the human who oversees an agentic stack not by managing the work but by managing the system that does the work — has a specific name in the Arco operating model: the Stewardship Model. We introduced it by name in Episode two and will cover it in full in Episode ten. What matters here is the relationship between the Stewardship Model and the Operational Drag mandate: the less-than-five-percent figure is not the target for the steward's productivity. It is the condition that makes the stewardship role possible. When Operational Drag is below five percent, the steward is not consumed by coordination work. They are free to do the architectural work that compounds the business over time. When Operational Drag is below five percent and MTTI is above seventy-two hours, the arithmetic of the business changes. The cost of the next unit of revenue approaches zero — because the logic is the engine of growth, not the labor. That is the structural path to the ten-to-one revenue-to-headcount advantage we established in Episode one.
The ten-to-one ratio is not a productivity aspiration. It is the arithmetical consequence of removing coordination overhead from the cost structure and replacing it with deterministic architecture. Operational Drag is the metric that tells you whether you are on that path. If Operational Drag is rising, the architecture is degrading — new human-in-the-loop dependencies are accreting, tier boundaries have shifted, or the business has grown into complexity the current design was not built to handle. If Operational Drag is holding below five percent, the architecture is sound. The mandate is not a ceiling to stay under. It is a signal to trust.
What is the Coordination Tax, and how does Arco eliminate it? The Coordination Tax is the overhead cost of human-to-human alignment — the meetings, approvals, status updates, and manual handoffs required to keep a traditionally structured business running. In legacy firms, this tax consumes twenty to thirty percent of operating budget. Arco eliminates it by replacing human-in-the-loop dependencies with deterministic agentic logic: agents share a single source of truth and execute handoffs at machine speed without requiring human coordination. The architectural target is an Operational Drag ratio below five percent of the total operational cycle — measured not as a productivity KPI but as a design constraint built in from the start.
Here is the test for overhead. If your costs are rising as fast as your revenue, you have not built a business. You have built a bureaucracy. The Coordination Tax is compounding. Operational Drag is high. The architecture is not holding — because the architecture was never designed to hold. It was designed around humans, and humans have coordination costs that scale with volume. Overhead is a choice. You make it at the design stage. Not when you hire the fourth approver on a report that should never have required one. Not when you run the quarterly restructuring programme to remove the layer of management that accumulated because no one designed it out. At the design stage — before the first workflow is built, before the first hire is made, before the first process is documented in a way that assumes a human will be present to run it.
Once the architecture is in place, removing it is expensive. The cost of reconstructing a system built under legacy assumptions at meaningful scale — the Rebuild Tax — is the compounding interest on every design shortcut taken in the name of moving fast. Episode four examines exactly that cost, and why it is the argument against building MVPs in the first place. The full written version of this argument is Memo number three — Overhead Is a Design Choice — on the blog at arcoventure.studio. Every term introduced across this three-episode arc is defined precisely in the Arco Lexicon, at arcoventure.studio/lexicon. Next week: why we do not build MVPs — and what we build instead. Growth should make a company more efficient, not more complicated. If your overhead is rising as fast as your revenue, you have not built a business — you have built a bureaucracy.
This has been Episode 3 of The Operator Log.