The Coordination Tax

A team of six doesn’t cost six salaries. It costs six salaries plus a tax nobody writes down.

The tax has a shape. Every person you add multiplies communication paths: n(n−1)/2 of them. Two people share one channel. Five share ten. Ten share forty-five. Twenty share a hundred and ninety. Fred Brooks worked out the consequence in The Mythical Man-Month decades ago: adding people to a late software project makes it later, because the new hire needs onboarding from the people who were producing, who now stop producing, while the new hire adds still more paths. And Conway showed where the tax ends up: in the product. You ship your org chart. The seams between teams become seams in the architecture.

Most of it is invisible on any budget line, which is why nobody writes it down. Onboarding. Context-sharing. Planning ceremonies. Status meetings. Review queues. Integration work. The drift between what you meant and what got built. Decision latency, all the hours spent waiting on someone else’s calendar.

We paid it anyway, and we were right to. One person couldn’t produce enough. The tax was the price of parallelism, and parallelism was the only way to scale output. That was a fair trade for seventy years, and I made it myself, over and over, at Intridea and after.

The strongest objection to this whole line of thinking is that the coordination cost is simply the cost of doing big things, and complaining about it is like complaining that bridges need engineers. For big things, that’s true. Nothing about a large, integrated, known build got cheaper to coordinate. If you’re constructing the deep system with the hard moat, you still need the team and you still pay the tax, and it’s still worth it.

What AI changed needs to be said precisely, because the imprecise version is hype. AI did not eliminate coordination. What it did was raise one person’s output ceiling enough that, for a class of work, you no longer need the parallelism. And if you don’t need the parallelism, you don’t pay the tax. The class of work is zero-to-one: discovery, prototyping, finding out whether the thing should exist at all. For that work, the tax used to be mandatory, because even exploration took more hands than one person had. Now it’s optional. That’s the entire change, and it’s enough.

One catch, honestly stated. Run multiple agents in parallel and a version of the tax comes back. But it’s the architectural slice, decomposition, interfaces, integration, divergent context, not the social slice. There are no meetings and no consensus-building. You orchestrate instead of meet. I’ve written separately about how the work has to be cut to make that true, and about where the orchestration itself starts to become management again.

For the first time in the history of this industry, paying the coordination tax is a choice. Not raw speed, not “10x developers,” that choice is the real shift. Which leaves the uncomfortable question for anyone staffing a project this year: are you paying the tax because the work demands it, or because paying it is the only way you’ve ever worked?