In 1973, Henry Mintzberg followed five senior managers around for a week and recorded what they actually did. His finding contradicted the textbook. Managers did not plan, organize, lead, and control in any recognizable sequence. He organized what they did into ten roles across three clusters: interpersonal (figurehead, leader, liaison), informational (monitor, disseminator, spokesperson), and decisional (entrepreneur, disturbance handler, resource allocator, negotiator). The manager, in Mintzberg's account, is not an authority figure seated at the apex of a command structure. The manager is a node at which information is translated, conflicts are resolved, and resources adjudicated.

Ronald Coase had already supplied the economic logic for why this node exists. His 1937 essay on the nature of the firm asked why firms exist at all, given that markets allocate resources efficiently and individual contracts can govern any exchange. The answer was that market transactions carry their own costs, including searching for counterparties, negotiating contracts, and enforcing them against opportunistic behavior. When a firm internalizes a function, it replaces those market costs with the cost of directing and coordinating people across specialized roles. Firms grow until the internal coordination cost of the next function exceeds what the market would charge. The management layer is not a power structure; it is what a firm pays to make specialists work together without renegotiating their relationship on every task.

Lyndall Urwick's student V.A. Graicunas formalized the scale problem in 1933 by counting the potential interactions between a manager and their reports. The formula is non-linear. A team of four generates 44 potential interactions by Graicunas's count; add a fifth person and that number rises to 100. This is why coordination overhead grows faster than headcount, why organizations add management layers as they scale, and why Bezos's two-pizza rule is less a preference than a theorem. The management layer does not thicken because organizations become bureaucratic. It thickens because coordination between specialists costs something, and that cost compounds.

The dissolving boundary

The Coasian case for a coordination layer rests on a specific condition. The specialists on either side of the manager cannot efficiently exchange information directly, so someone must translate between them. An engineer who cannot assess user needs must wait for a PM to convert user research into requirements. A designer who cannot evaluate technical feasibility must wait for an engineering review before committing to an approach. A PM who cannot build even a rough prototype must rely on an engineer to determine whether an idea is buildable in principle. The manager absorbs these translation costs, and what the management layer charges the organization is the sum of that translation work across every specialist boundary.

AI compresses the skill gap between adjacent specializations with enough speed to change the organizational calculus. A product manager who can use a code-generation tool to produce a working prototype does not need an engineering counterpart to validate whether the idea is buildable. The prototype is the validation. A designer who can generate functional front-end code does not need a handoff meeting with an engineer to test whether a layout is technically feasible. The adjacent knowledge that previously required a specialist, and a coordinator to mediate between them, is now accessible without either. When the translation cost at each specialist boundary approaches zero, the manager whose job was to perform that translation loses the function that justified the role.

AI lowers the cost of a single person holding enough of each specialization to do the job without a coordinator between them. The PM, the designer, and the engineer are not disappearing as functions; the specialization that required a dedicated translation layer between them is. The product builder who moves fluidly across product thinking, design judgment, and basic implementation appeared first at Apple, where Kocienda describes engineers who understood design and designers who understood code. AI makes that combination achievable at lower organizational density. A three-person team can now carry the creative integration that previously required ten, plus the coordination infrastructure to hold ten together.

What disappears

The coordination tax is the set of roles whose primary function was translation between specialists who could not coordinate directly. It includes the PM who spent most of their time writing requirements documents so engineers could build what designers had specified. It includes the program manager whose job was to track whether those handoffs had happened and escalate when they had not. It includes the middle management layer that existed to absorb coordination overhead. That layer grew because Graicunas was right about interaction complexity, and it will shrink because the interactions it was managing are going away.

None of these roles disappear because AI is doing their job. They disappear because the specialist boundary they were bridging is dissolving. A coordinator between two specialists is necessary only as long as neither specialist can do enough of the other's work to communicate directly. When that threshold shifts, the coordinator becomes a cost the organization is paying for a problem it no longer has. Apple's Directly Responsible Individual (DRI) and Amazon's Single-Threaded Owner (STO) were early evidence that organizations understood this intuitively. Both attempted to minimize the coordination surface by concentrating authority within a specialist structure that persisted. AI allows a more radical move by eliminating the specialization that made authority concentration necessary in the first place.

What the mechanism predicts

What dissolves when the coordination layer goes is the translation function, not the informational or decisional ones. Mintzberg's informational cluster covers the monitor, disseminator, and spokesperson roles. These are not coordination functions in the sense of bridging specialists but intelligence-routing functions, and they intensify when teams are small and autonomous because no one inside the build cycle has time to look outward. The decisional cluster is different. Disturbance handling, resource allocation under constraint, and negotiation between competing interests are not translation work. They survive the dissolution of specialist boundaries intact because small autonomous teams in competitive organizations generate exactly these conflicts regardless of how the work is organized.

A three-person product team that can see all of its own work cannot see the competitive environment it operates in. The team knows what it is building. It does not know where that build sits in the organization's broader priorities, what the market is doing that should change what the team considers worth building next, or who decides when two small teams are competing for the same scarce resource. On a large team with a dedicated PM, some of that external scanning happens automatically, and the PM holds enough organizational authority to settle priority disputes without escalating each one to leadership. On a small, vertically integrated team, both of those functions go when the coordination layer does. The need for them does not.

The role that replaces the coordinator is the context-setter, a manager responsible for reading the competitive environment the small team cannot observe from inside the build cycle and arbitrating the resource and priority conflicts it cannot resolve without organizational authority. One context-setter can serve several small teams because the role scales with strategic scope rather than with team-level coordination overhead. The STO was an early version of this role, but it was operating in a world where the specialist structure still existed and the STO still needed to manage it. The context-setter provides the strategic frame within which a vertically integrated team can make autonomous decisions without waiting for approval on each one. The team determines how to build; the context-setter ensures the team knows why the build matters and what evidence would make them reconsider it.

Job requirements

The context-setter requires capabilities the coordinator role never demanded. Competitive awareness is the primary one, meaning knowledge of where the market is moving, what adjacent products are doing, and where the organization's actual priorities diverge from its stated ones. Editorial authority is the second requirement, which Kocienda describes as the willingness to tell a team that what it built is wrong, without a committee to defer to or a metric to hide behind. The third is a willingness to operate at the boundary of the organization, in strategy conversations and customer interactions that never produce a ticket, rather than inside the sprint cycle. The fourth is the authority and willingness to adjudicate between competing teams when priorities conflict, resources are constrained, and the right answer is not in a framework. The coordinator lived inside the process. The context-setter lives at its edge.

The DRI and STO pointed at this structure before the organizational density was available to achieve it fully. Amazon's STO tried to create a context-setter but still needed to manage coordination overhead because the teams beneath were still specialist teams. Apple's DRI named accountability but still assumed a horizontal function structure underneath. What AI enables is the completion of the argument those frameworks started. The specialist boundary can now dissolve rather than be held in place by concentrated authority. When it dissolves, the only remaining managerial function is context-setting, meaning the work of reading the competitive environment the team cannot observe from inside the build cycle and resolving the conflicts the team cannot resolve without someone who holds organizational authority.

References

Mintzberg, H. (1973). The Nature of Managerial Work. New York: Harper & Row.

Coase, R.H. (1937). The Nature of the Firm. Economica, 4(16), 386–405. https://doi.org/10.1111/j.1468-0335.1937.tb00002.x

Graicunas, V.A. (1933). Relationship in Organisation. In L. Gulick & L.F. Urwick (Eds.), Papers on the Science of Administration. New York: Columbia University, Institute of Public Administration (published 1937).

Kocienda, K. (2018). Creative Selection: Inside Apple's Design Process During the Golden Age of Steve Jobs. New York: St. Martin's Press.

Lashinsky, A. (2012). Inside Apple: How America's Most Admired—and Secretive—Company Really Works. New York: Business Plus. (DRI concept first reported in Fortune, May 2011.)

Bryar, C. & Carr, B. (2021). Working Backwards: Insights, Stories, and Secrets from Inside Amazon. New York: St. Martin's Press.