Abstract

AI usually delivers its first gains locally. One person writes faster, one team drafts better, one function resolves more routine work. That is real value. But local gain is not the same thing as system gain. Once agents are introduced into a live organization, the first cost is often coordination: more review, more exception handling, more trust calibration, more handoff management, and more work spent making the new layer legible to the rest of the business.

This is the coordination tax. It is not a flaw in AI. It is the price of introducing a new actor into a system that was built around a different division of labour. The tax is temporary when the organization redesigns the workflow well. It becomes permanent when leadership mistakes local speed for system coherence.

Reworking Order uses this idea to explain a practical pattern: the first wave of AI adoption often makes work feel more complicated before it makes it simpler. That does not mean the change is bad. It means the organization is paying for a new layer of coordination and has to decide whether the payoff is worth it.

Local Wins, System Costs

The easiest mistake in AI adoption is to judge the system by the best individual user experience. That is not enough.

If one person can draft faster, that is a gain. If one analyst can summarize more information, that is a gain. If one operator can move a task through a queue with less effort, that is a gain. But the organization does not operate on isolated wins. It operates on the interactions between those wins.

As soon as AI enters a live workflow, people have to ask new questions:

  • Is the output trustworthy enough to pass to the next step?
  • Who checks it?
  • What counts as a failure?
  • What gets escalated?
  • What happens when the agent is confident but wrong?
  • How do we preserve accountability when the pace increases?

Those questions create overhead. Some of that overhead is visible, but most of it is not. It shows up as more comments in review threads, more checks in approval chains, more time spent explaining system behavior, and more attention given to edge cases that used to be handled informally.

That is the coordination tax.

It is especially visible in the gap between individual productivity and team-level coherence. A worker may finish faster, but the team may not move faster if the output now needs more validation. A department may produce more drafts, but the downstream group may absorb the complexity. A manager may celebrate speed, while the actual organization pays for the new layer through friction elsewhere.

This is why AI adoption can feel deceptively easy at first and frustrating shortly after. The tool is doing something useful, but the business is also learning how to absorb the tool.

Where The Tax Shows Up

The tax appears in a few repeatable places.

First, in review. When AI can produce plausible output quickly, teams often respond by adding review. That is rational. But review itself consumes time, and that time has to come from somewhere.

Second, in exception handling. AI is strongest when the world fits the pattern. Real organizations are full of exceptions, special cases, and local conditions. Every exception adds a little coordination burden, and those burdens accumulate.

Third, in trust calibration. People need to learn what the agent is good at, where it fails, and what it should never be allowed to decide on its own. That learning curve is real work.

Fourth, in standardization. Once an agent is used widely, teams want a shared prompt library, a policy, a template, a review gate, or a quality bar. Those are useful things, but they are also structural additions. They make the system more coherent, but only after they make it heavier.

Fifth, in ownership. If everyone can rely on the agent, it becomes harder to tell who actually owns the outcome. That ambiguity is manageable at first and expensive later.

The coordination tax is not just a cost of the technology. It is a cost of the transition between two ways of working.

That is why the tax is most visible when a team tries to keep the old workflow intact while adding a new actor into it. The old workflow was optimized for humans. The new one needs room for the agent, room for review, and room for different kinds of handoff. If none of that changes, the business quietly starts paying in friction.

The research supports this reading. Work on generative AI in firms shows that the earliest gains often appear in individually adjustable tasks rather than in coordinated team structures. Other research on AI adoption and system-wide change makes the same point more formally: the organization has to redesign around the technology if it wants the gains to compound. AI does not eliminate coordination. It changes what coordination costs and where the costs land.

When The Tax Is Worth Paying

The coordination tax is not a reason to avoid AI. It is a reason to be precise about where the payoff exists.

If the work is repetitive, bounded, and easy to review, the tax may be small and the payoff immediate. If the work is ambiguous, cross-functional, or high-stakes, the tax may be larger, but the strategic upside may still justify it if the redesign is done well.

The practical question is whether the coordination burden is temporary and productive or permanent and wasteful.

Temporary and productive looks like this:

  • more review now, less manual repetition later
  • more explanation now, clearer workflows later
  • more calibration now, more autonomy later
  • more design work now, less chaos later

Permanent and wasteful looks like this:

  • more review without better decisions
  • more prompts without clearer ownership
  • more agent activity without more coherence
  • more output without more accountability

The distinction matters because a lot of teams confuse busyness with progress. A new tool can create a lot of motion. The question is whether that motion reduces the total cost of getting work done.

That is why Reworking Order talks about the build vs deliver gradient. Some coordination tax is worth paying because it buys reuse, leverage, and clarity. Some is not. The work is deciding which is which.

The best organizations will not be the ones with the lowest immediate friction. They will be the ones that absorb the right amount of friction in order to establish a better operating model.

The Tax Is Also A Signal

The coordination tax tells you where the redesign has not happened yet.

If every AI use case creates review bottlenecks, the workflow probably still assumes too much human uncertainty upstream. If every agent output requires heavy interpretation, the standards are not clear enough. If people keep redoing the same correction work, the system has not yet learned what should be made durable.

In that sense, the tax is diagnostic. It shows where the organization still lives in the old order.

That makes it useful. It tells leaders not only that something is slowing down, but where the new division of labour has not been settled. It also tells them what kind of redesign is needed:

  • tighter scope
  • clearer approval paths
  • better exception rules
  • stronger prompts and templates
  • better review gates
  • explicit human ownership of consequential decisions

AI is not removing coordination from work. It is making coordination more visible. The businesses that understand that will use the tax as a guide. The ones that do not will treat it as a nuisance and miss the structural change underneath it.

Diagram

timesystem effectcoordination burdenlocal speed upsystem payoffshort-term: more review, trust calibration, handoffsReworking Order
Local productivity rises first, coordination burden rises with it, and system payoff only arrives later if the workflow is redesigned well enough to absorb the new load.

Sources