Legacy Liability. Why Incumbents Can't Adapt.

The Operator Log, Episode six. What We Observe. Legacy Liability. Why Incumbents Can't Adapt. The structures that built the incumbent are the same structures preventing it from changing.

Last week we covered how Arco identifies a target market. We ended with a specific promise: why the incumbent in that market cannot respond even when they see us coming. That is this episode. The conventional defence of the large incumbent is scale. Distribution, brand, regulatory relationships, institutional knowledge — all real and all genuinely difficult for a new entrant to replicate. They are also, precisely, what makes the incumbent impossible to reform from the inside. The structures that built the incumbent are the same structures preventing it from changing. Size is no longer a moat. It is a weight. Legacy Liability is the structural condition of a business that has grown too dependent on human-centric coordination to rebuild itself without dismantling the organisation in the process. The incumbent cannot eliminate its Coordination Tax without eliminating the departments built around it — and no board votes for that. This is The Operator Log.

The case for the incumbent is not weak. A large logistics firm has carrier relationships built over decades. A back-office compliance operation has regulatory clearances, audited processes, and a client list that a new entrant cannot inherit. A professional services brokerage has institutional knowledge distributed across a team of experienced coordinators whose understanding of the market took years to accumulate. These are genuine advantages. They are not nothing. They are also precisely what makes autonomous reconstruction impossible from the inside. The primary obstacle for a large enterprise attempting autonomous transformation is not technology. It is the organisational chart. A traditional firm is a collection of human handoffs, approval chains, and middle-management layers designed to coordinate those handoffs. Each of those layers represents a budget line, a headcount, a set of job titles, and a reporting structure. Removing them is not an engineering decision. It is a political one — and boards that have spent years building the organisation have no structural incentive to dismantle it in the name of efficiency that would not accrue to the people who authorised the dismantling. When enterprises attempt to implement agentic AI into this structure, they make a consistent and observable category error: they fit the agent into an existing human role. A coordinator gets an AI assistant. A reviewer gets a document scanner. A manager gets an AI dashboard. The agent is inserted into the workflow at the point where the human previously acted — and immediately throttled by the slowest links in the chain that surround it. The approval gate before the agent is still human. The escalation path above the agent is still human. The handoff following the agent's output is still human. The result is what the memo for this episode calls AI-assisted inefficiency: faster inputs meeting the same bottlenecks. The incumbent pays for both the expensive human overhead and the new technology, without achieving the structural decoupling of headcount from revenue that autonomous architecture requires. The Coordination Tax persists — because the human-in-the-loop dependencies that generate it persist. Adding a faster tool to a slow system produces a faster slow system. This pattern is not unique to any single industry. It is the observed outcome of every AI transformation programme that layers technology onto an existing human architecture rather than redesigning the architecture itself. The consultants identify the inefficiency. The roadmap is approved. The pilots run. The results show incremental improvement. The structural cost does not move. The programme is declared a partial success. The Coordination Tax continues to run. Corporations optimise for continuity. Arco optimises for replacement. Those are not two strategies in the same game. They are two different games entirely.

The Coordination Tax in a large enterprise is not an estimate. It has been measured, repeatedly, across industries and geographies, over more than a decade. The data is consistent enough to be treated as a structural observation rather than a research finding. The McKinsey Global Institute found that the average knowledge worker spends 28% of the working week managing email and a further 20% searching for information and tracking down colleagues. That is 48% of the working week consumed before a single unit of productive output is possible. Not in underperforming companies. In average ones. Hamel and Zanini, writing in Harvard Business Review, quantified the budget consequence: excess management overhead in large firms consumes up to 30% of operating costs — structural spend that produces no output, only alignment. Their broader analysis put the cost to the US economy at more than three trillion dollars annually in lost productive output. This is not a rounding error. It is the arithmetic of organisations designed to coordinate humans rather than to produce output. Miro's 2025 Momentum at Work research confirmed the pattern in present-day operating conditions: 79% of knowledge workers cite constant emails and messages as their primary maintenance burden, and spend three hours on coordination for every one hour of work that generates output. The ratio has not improved with the introduction of AI tools into the enterprise. It has held. Microsoft's 2024 Work Trend Index added a fourth measurement from a different angle: 60% of time in productivity tools — the software explicitly built to improve output — is consumed by email, chats, and meetings rather than by work that produces anything. The tools built to solve the problem have become vectors for the problem. Taken together, these four observations describe the same structural condition from four different measurement points: the Coordination Tax in a large enterprise is not 5% or 10% of operating capacity. It is between 30% and 50% of everything the organisation does. And it has not responded to thirty years of productivity software, digital transformation programmes, or now AI tools — because none of those interventions removed the human-in-the-loop dependencies that generate it. To make this concrete at the operational level: consider a logistics firm with a team of freight coordinators. Their working day consists of matching freight with carriers, confirming availability, resolving scheduling exceptions, and updating status across multiple parties. The workflow was designed for humans — the data sits in fragmented systems that do not talk to each other, every exception requires a judgment call, and the institutional knowledge of which carrier to call for which lane lives in the coordinator's head rather than in any accessible system. An AI assistant speeds up the email. It does not remove the email. The exceptions still require a human. The carrier relationships still require a human. The data fragmentation means the agent cannot read the inputs it would need to act deterministically. The Coordination Tax persists in full, because the structural conditions that generate it — fragmented data, human-in-the-loop decisions, institutional knowledge stored in people rather than systems — have not been addressed. Adding faster tools to an architecture that requires human handoffs produces faster human handoffs. The tax runs at the same rate. An incumbent cannot delete this tax without deleting the departments built around it. The middle-management layer that coordinates the coordinators is not an inefficiency that can be quietly removed — it is a structural feature of how the business was designed. Every attempt at digital transformation runs into this wall: the technology is straightforward; the organisational surgery required to implement it properly is not. They are trapped not by incompetence but by the accumulated weight of every architectural decision made in favour of continuity over reconstruction.

The question that follows from the evidence is always some version of this: if the Coordination Tax is so well documented, why have enterprises not eliminated it? They have had the data for more than a decade. They have the consulting budgets. They have the stated will to transform. Why hasn't it happened? The answer is structural, not motivational. Eliminating the Coordination Tax requires eliminating the human-in-the-loop dependencies that generate it. In a large enterprise, those dependencies are the organisation. The approval chains are the management structure. The handoff meetings are the coordination mechanism between departments that cannot read each other's state without a human bridge. The status updates are the communication layer that keeps a distributed workforce aligned. Remove the dependencies, and you remove the departments built around them. No board votes for that — not because they lack vision, but because the people who would authorise the reconstruction are the same people whose roles the reconstruction would eliminate. This is why the AI transformation programmes that receive significant investment produce incremental results. The consulting firm maps the inefficiency accurately. The pilot demonstrates the technical feasibility. The business case is approved. Then the implementation runs into the organisation. The agents are deployed alongside the humans rather than replacing them. The approval gates remain because removing them requires a decision no one will make. The Coordination Tax continues to run on the human workflow while the AI infrastructure is paid for in parallel. The incumbent ends up with two cost structures and the productivity gain of one. The distinction we drew in Episode 01 applies directly here. An automated business optimises existing workflows — makes the humans faster, reduces the friction within the existing architecture. An autonomous business replaces the architecture entirely. The incumbent attempting AI transformation is, at best, moving from one type of automated business to a slightly more automated one. It is not moving toward autonomy, because autonomy requires a clean-sheet design — and a clean-sheet design, for an incumbent, means dismantling the organisation that the existing design produced. That cost is the Rebuild Tax applied at the scale of a large enterprise. It is never paid. Arco's response to this observation is not a consulting engagement. It is the Arco model itself. We do not enter markets to transform the incumbent. We enter markets to replace the function the incumbent performs — using an architecture that was never built around human handoffs and therefore does not carry the structural debt that prevents the incumbent from changing. The markets Arco targets are the ones where incumbents have been accumulating Legacy Liability for decades. The longer a firm has run on human-centric coordination, the deeper the structural debt — and the wider the gap between what the incumbent charges for its service and what an autonomous competitor needs to charge to generate the same margin at Arco's Operational Drag target. That gap is the Operational Arbitrage. The incumbent created it. Arco captures it. Not by being smarter or faster or better funded. By having built correctly from the start — which is only possible because we never had legacy to begin with. Arco does not wait for incumbents to digitally transform. We wait for their Coordination Tax to become so high that they can no longer compete on price or speed. Then we move in — not with a better product, but with a structurally superior cost base. That cost base is the consequence of five episodes of architectural decisions that all pointed in the same direction: autonomous by design, operational by default, and built without a single human-in-the-loop dependency that did not earn its place. Episode 07 examines why the studio model is the right vehicle for this approach — and why the founder-led model, for all its strengths, carries a version of the same structural constraint that limits the incumbent.

What is Legacy Liability and why can't incumbents eliminate it? Legacy Liability is the structural condition of a business that has grown too dependent on human-centric coordination to rebuild itself without dismantling the organisation. The Coordination Tax embedded in a large enterprise — measured at 28–48% of working time by McKinsey Global Institute research, and up to 30% of operating costs by Harvard Business Review analysis — cannot be deleted without deleting the management layers that define the company. Incumbents attempting AI transformation consistently produce AI-assisted inefficiency: faster inputs meeting the same human bottlenecks. Autonomy requires clean-sheet design. A business built on human handoffs cannot become autonomous by adding agents to it. It can only be replaced by a business that was never built that way.

Here is the verdict on the incumbent. The advantages that built it — scale, distribution, institutional knowledge, regulatory relationships — are real. They are also precisely why the reconstruction that autonomous architecture requires is unavailable from the inside. The same size that made the incumbent dominant makes it structurally resistant to change. Every consultant hired to accelerate the transformation runs into the same wall: the organisation cannot be made autonomous without first being made smaller, and no board votes for that. The data from four independent research programmes — McKinsey, Harvard Business Review, Miro, and Microsoft — all describe the same condition: between a third and a half of everything a large knowledge-work enterprise does is consumed by the organisation managing itself rather than by work that produces output for customers. That proportion has not declined with thirty years of productivity software or with the current wave of AI tool deployment. It has held — because the tools have been added to the human architecture rather than replacing it. Arco does not compete with incumbents on their terms. We build businesses on the structural cost base that incumbents cannot reach — because reaching it would require them to dismantle the organisation they spent thirty years building. The Operational Arbitrage is not something we created. It is something they accumulated. We are simply the operator positioned to capture it. The full written version of this argument is Memo #06 — Legacy Liability — on the blog at arcoventure.studio. Every term introduced across this six-episode arc is defined precisely in the Arco Lexicon, at arcoventure.studio/lexicon. Next week: why the studio model is the right vehicle for autonomous business — and why the founder-led approach, for all its strengths, carries its own version of the structural constraint we examined today. You cannot fix a legacy business by adding agents to it. You can only win by building a business that does not have legacy to begin with.

This has been Episode six of The Operator Log.