The coordination tax is the only tax worth cutting

Add the second person to a one-person company and something strange happens to the math. The work that needs doing didn’t double, but the day got longer. Now there’s a standup. A handoff. A “did you see my message?” A second copy of the context, living in a second head, slowly drifting out of sync with the first. The new person was hired to add capacity. Half of what they added was the cost of staying coordinated with the person who hired them.

Nobody put that on an invoice. It’s the most expensive line item in the company and it has never once appeared in the budget.

Most of what a growing company spends isn’t on doing the work, it’s on keeping humans in sync. That’s the coordination tax, and the rest of this post is about why it scales the way it does, why no headcount can pay it down, and what changes when the people staying in sync aren’t people.

The tax nobody itemizes

Every company has a budget. Salaries, software, rent, the cloud bill. Founders track all of it to the dollar.

Then there’s the spending that never shows up. The hour the engineer spent reconstructing context for the new hire. The proposal that went out late because three people each owned one paragraph and the document waited on whoever was slowest to reply. The decision made twice because the first decision lived in a meeting nobody wrote down. The duplicated work, the dropped ball, the thing that fell between two roles because each assumed the other had it.

The biggest cost in a growing company is invisible: it’s everyone keeping everyone else up to date.

None of that is laziness, and none of it is fixable by hiring better people. It’s the structural cost of running work across more than one mind. I’ll call it the coordination tax: everything you spend not on the work itself, but on keeping the humans doing the work aligned about it.

The naive view is that this tax is small, a few meetings, some Slack, the price of being a team. That feels right when you’re three people in a room. It stops being right fast, and the reason is geometry.

Why the tax grows faster than the company

Here’s the part that surprises founders. Coordination cost doesn’t grow with the number of people. It grows with the number of connections between people.

Add the second person and you have one relationship to keep in sync. Add the third and you have three. The fourth brings the total to six, the fifth to ten. The headcount went up by one each time; the number of pairs that have to stay aligned went up by two, then three, then four. This is just counting, the pairs in a group of n people come to n times (n minus 1), divided by two, and it’s the quiet engine behind why a fifteen-person company can feel slower than a five-person one despite tripling the hands on deck.

Software people have known this for fifty years and have a name for it. Brooks’s Law, from The Mythical Man-Month in 1975, says adding people to a late project makes it later, precisely because each new person multiplies the communication overhead (Wikipedia: Brooks’s Law). The work the new person does is linear. The coordination they trigger is closer to quadratic. Past some point, the next hire spends more time getting and staying in sync than they spend producing, and so does everyone who now has to sync with them.

This is the trap that makes scaling feel like wading into deeper water. You add capacity and the company speeds up, for a while. Then you add more and it doesn’t, and you can’t see why, because the thing slowing you down isn’t on any dashboard. It’s the tax, compounding in the connections. More of what the growing company spends goes not to doing the work but to keeping its humans in sync, and that share rises with every hire.

So companies do the only thing the org chart allows.

How we’ve always tried to pay it down, and why it never clears

The classic move is to manage the coordination instead of removing it. You add structure: a manager whose job is to route information between people. A weekly meeting to re-sync everyone. A project tool where status is supposed to live. A process document so the handoff is “repeatable.”

Watch what each of these actually is. A manager is a human router for context. A standup is a scheduled re-sync because the context drifted overnight. A status field is a place to write down what one person knows so another can read it. Every one of these is a payment on the coordination tax, and every one of them is also a new node that has to be kept in sync, which is to say, a new line of tax.

That’s the cruel part. The tools we reach for to reduce coordination overhead are themselves coordination overhead. The manager has to be briefed. The meeting has to be prepared for. The status field has to be updated, and it’s stale the moment someone forgets, which is always, because updating it is pure tax with no felt reward.

The deeper reason none of it clears the bill: every human carries their own private copy of the context, in their own head, and those copies drift the instant they’re made. The proposal in your head and the proposal in mine were the same yesterday and aren’t today. All of coordination, every meeting, every message, every status update, is the ongoing, never-finished work of reconciling copies that won’t stop diverging. You’re not paying the tax down. You’re renting alignment, by the hour, forever.

You don’t pay the coordination tax once. You pay it every time two private copies of the truth drift apart.

Hold onto that diagnosis, because it points straight at the fix. The tax isn’t caused by having many minds. It’s caused by every mind holding a separate copy of the truth. Remove the separation and the tax doesn’t get smaller. It nearly disappears.

The thing that changes when the workers share one memory

So picture work done not by people each holding a private copy, but by agents reading and writing one shared memory.

When two people collaborate, there are two copies of the context and a channel between them that has to carry every update, perfectly, forever, and it never does. When two agents collaborate on one company brain, there is one copy. The thing the first agent learned at 9am is not something it has to tell the second agent. The second agent reads the same memory. There is no message to send, no meeting to hold, no status to update, because there is nothing to keep in sync, the sync is the substrate.

This is the move that breaks the geometry. The reason coordination cost grew with the connections between people was that each connection was a fragile channel between two private copies. Collapse the copies into one shared memory and the connections stop being channels that can drift, they become reads against the same source. The number of agents can climb, and the coordination cost doesn’t climb behind it, because no agent is maintaining a separate truth that has to be reconciled with anyone else’s.

Think about what that deletes from a normal week. The handoff is gone, there’s nothing to hand off when both workers already see the same context. The “did you get my message?” is gone, there’s no message, just a shared state that’s already current. The re-sync meeting is gone, nothing drifted overnight, because nothing was holding a private copy to drift. The duplicated decision is gone, the second agent sees that the first already decided. What’s left is the work itself, paid for once, by whichever agent does it, with no tax skimmed off the top for keeping everyone aligned.

Most of what a growing company spends isn’t on doing the work, it’s on keeping humans in sync. Agents on one memory pay almost none of it.

The objection worth taking seriously

The honest pushback is this: isn’t shared memory just a database, and haven’t companies always had databases?

It’s a fair challenge, and the difference is what the workers do with it. A database is a place humans read from and write to when they remember to, in their own format, on their own schedule, which means it’s just another copy that drifts from the copies in their heads, one more thing to keep in sync. The CRM that’s always out of date is exactly this: a shared store that nobody’s private context actually matches.

The change isn’t that the memory exists. It’s that the workers live in it. An agent doesn’t keep a private copy and occasionally update the shared one; the shared memory is the only memory it has. It reads its full context from there at the start of every task and writes everything it learns back before the task ends. There’s no second copy in a head to drift, because there’s no head. The store stops being a place people sync to and becomes the place the work actually happens. That’s the line a plain database never crossed.

The turn: the part of the company that was never a tax

Here’s what’s actually worth saying, and it isn’t about software.

If you strip the coordination tax out of a company, you don’t just get a faster company. You find out how much of everyone’s day was tax and how little was the thing you actually hired them for. The manager who spent their week routing context can spend it on judgment. The person who spent half their hours staying in sync can spend them on the work only they can do. The tax was never the talent. It was the friction wrapped around the talent, and most careers have been spent paying it.

Because here’s the thing the shared memory can’t hold: what’s worth coordinating toward. An agent that reads the same context as every other agent will never argue about who knew what, but it can’t tell you which problem deserves the company’s attention, what “good” means for the people you serve, or which bet is worth the year. That judgment doesn’t drift between copies, because it was never a copy. It’s the one thing in the company that was never a coordination cost, because it’s the thing all the coordination was for.

Cut the tax, and what’s left is the only work that was ever worth a human’s full attention. The aligning was never the job. Deciding what to align around was.

---

That’s what we’re building at Apollo Space: a company where the workers share one memory instead of a hundred drifting copies, so the tax of staying in sync stops eating the work, and the only thing left for you to spend your day on is the part no shared memory will ever decide for you. If your company feels slower the more people you add, you’ve already met the tax. We think it’s the one worth cutting first.