The coordination tax, and the person quietly paying it
By Ismail El Houcheimi
Somewhere between roughly five and fifteen people, most teams acquire a role nobody hired for and nobody named. One person becomes the human glue. They chase loose ends. They notice a task has no owner and find one. They remember that the thing you are about to start depends on something still blocked. They send the message that unsticks a two-week stall.
They are usually good at it, often the reason things ship, and frequently the last person anyone thinks to protect. What I want to argue is that this is not a personality trait or a management gap. It is a predictable structural cost, and treating it as somebody being helpful is how it becomes permanent.
What the work actually consists of
Coordination overhead is easier to dismiss than to describe, so it is worth being specific. Almost all of it falls into four buckets.
- Reconciling state. Determining what is actually true right now, because the board says one thing, the chat thread says another, and the person who knows is in a meeting.
- Assigning ownership. Noticing work that exists but belongs to nobody, and either taking it or finding whose it is, which usually means a conversation, not an edit.
- Tracking blockers. Holding in your head which things are waiting on which people, and how long they have been waiting, because nothing else is holding it.
- Re-transmitting priority. Restating what matters most this week, because a decision made once in one place does not propagate to everywhere the work lives.
None of these produce artifacts. That is the defining feature. At the end of a day spent entirely on this, there is nothing to point at, which is why the person doing it often feels they got nothing done, and why it never appears in a review.
Why it forms, reliably, at a certain size
The mechanism is not mysterious. Coordination cost grows with the number of connections between people, not the number of people, and connections grow much faster than headcount. Three people have three pairs. Ten people have forty-five. Nothing about anyone becoming lazier explains the change.
What makes it acute rather than merely annoying is fragmentation. In a small team the state of work fits in one place, often one person head. Past a certain size it does not: commitments get made in meetings, tracked in a tool, discussed in chat, and completed in a repository. No single view is authoritative. Somebody has to be the join across those systems, and since no system volunteers, a person does it.
Why leaving it alone is expensive
It is tempting to accept this arrangement, because it works. Someone competent is absorbing the friction and the work is getting out. Three things make that a bad trade over any real time horizon.
It concentrates knowledge in one head
The glue person accumulates an unwritten model of how everything connects. When they are on holiday the team does not slow down slightly; it loses its map. This is a single point of failure that nobody has classified as one.
It burns out the wrong person
The work is interrupt-driven, invisible, and unrewarded, which is close to a recipe for attrition. The person most responsible for things working is the one most likely to leave, and the exit interview will not mention coordination.
It scales linearly with headcount
Every hire adds coordination load. Because the cost is absorbed rather than measured, this shows up as mysterious slowdown (more people, no more output) and gets misdiagnosed as a hiring-quality or motivation problem.
What reduces it, and what just moves it
The instinct is more process: another standup, a stricter template, a weekly sync. Some of that helps. Much of it converts coordination overhead into meeting overhead, which is the same cost paid by more people at once.
The things that genuinely reduce it share a property: they remove the need for a human to be the join between systems.
- Make ownership structural rather than remembered. Work that cannot exist without an owner cannot generate the "whose is this?" conversation.
- Make blocked state explicit and visible, including how long it has been blocked. Most stalls persist because nobody is counting, not because nobody cares.
- Reduce the number of places authoritative state lives. Every additional system is another edge a person has to traverse manually.
- Make drift surface itself. A project with a fixed date and no activity for three weeks is a query, not something that should depend on a person noticing.
- Separate capture from commitment, so that recording something does not require deciding who owns it in the same moment, the friction that produces unowned work in the first place. The same split is what keeps a task backlog from becoming an archive.
Reducing this overhead is the problem we are building Serena for teams, so discount the framing as you see fit. The part I would defend regardless: name the cost and find out who is paying it. A team that cannot say who its glue person is usually has one, and has not noticed what it depends on. Surfacing the unowned task, the stale blocker, and the priority that did not propagate is the part Serena AI takes on.
Key takeaways
- Coordination overhead is four things: reconciling state, assigning ownership, tracking blockers, re-transmitting priority.
- It produces no artifacts, which is why it is invisible in reviews and metrics.
- It grows with connections between people, not headcount, and bites hardest when authoritative state is fragmented.
- Left alone it concentrates knowledge in one head, burns out the person holding it, and makes new hires look unproductive.
- Fixes remove the human join between systems; more process usually just relocates the cost into meetings.
Frequently asked questions
Is the glue person just doing a manager job?
Sometimes, but frequently they have no authority: a senior engineer, an early employee, an ops hire who absorbed it. That mismatch is part of the problem: they carry responsibility for things working without the standing to change how the team operates.
How do you measure something that produces no artifacts?
Indirectly, and roughly is enough. Count how many items in a week had no owner when work started, how long the longest-blocked item sat before anyone raised it, and how many status questions got answered by a person rather than a view. You are looking for a trend, not a number.
Can this be solved by hiring a program manager?
It can be made survivable, which is worth something. But it formalizes the tax rather than removing it, and it adds a person whose absence now degrades the team. Reasonable at a certain scale; a poor first move at fifteen people.
What if our team genuinely needs the flexibility of loose process?
Then reduce the number of places state lives rather than adding ceremony. Most coordination pain comes from fragmentation, not from insufficient rules, and consolidating where the truth lives costs a team almost none of its flexibility.