The bus factor is two

Six weeks in, every joint project runs on two people — one on each side. Everything that matters is in their heads, and neither company has a copy.

Six weeks into any decent joint project, ask both companies a specific question — why the integration authenticates the way it does, or who at the client actually has to approve a design change — and you will be told to go and ask a particular person. On each side there is exactly one of them.

Nobody appointed these two. There’s no line in the contract naming them. They emerged, the way a footpath emerges across a lawn, because the work needed a route and they were standing in the right place.

They are also, at this point, the project.

Every joint project grows a shadow org chart

The concentration happens fast, and for good reasons. Cross-company communication is expensive — you have to figure out who to ask, phrase things carefully because they’re a client or a supplier, and wait through a time zone. So people stop doing it directly. An Atlas engineer with a question for Meridian asks the Atlas delivery lead, who asks her counterpart, who asks their engineer, and the answer comes back the same way.

It’s efficient. It’s also a single point of failure that shows up on no risk register, because the two people in the middle look like coordination overhead rather than infrastructure.

All cross-company threads routed through two peopleTwo company territories, each holding five people. Every person connects only to one hub inside their own company, and the two hubs connect to each other across the dashed company line by a single thick link. All joint context flows through those two people.the company lineAtlas StudioMeridian CorpengdesignqaaccountenglegalfinanceengsecurityproductATMEthe entire joint recordheld by two people, not two companiesdelivery leadhere since week oneproject managerthe only one who read the spec
The shadow org chartTen people are on this project. Two of them are load-bearing. Cut either node and the other company is suddenly talking to a stranger about a six-month history it can't produce.

The relationship isn't between two companies. It's between two people — and nobody chose that.

What actually walks out the door

The tidy version of this risk is “documentation.” Write things down, the thinking goes, and the exposure goes away. Then you look at what’s actually in those two heads and notice that almost none of it belongs in a document either company owns.

Some of this is in an email somewhere. That is not the same as being findable, and it definitely isn’t the same as being trusted — a six-month-old thread on a decision that may have been superseded twice is worse than nothing, because someone will act on it.

Then one of them leaves

It doesn’t take a resignation. Two weeks of annual leave produces a small version of the same event, which is why joint projects mysteriously lose momentum every August.

The real thing looks like this. The delivery lead gives four weeks’ notice. Atlas does a proper internal handover — accounts, files, the tracker, a long call. What can’t be handed over is the half of the context that lives on the other side of the company line, because Atlas has no copy of it. The new lead inherits a relationship, not a record.

So they start asking. Most of the questions have been answered before, and the answers are with Meridian, who now spends five or six weeks re-explaining a project they thought was understood. That cost lands on the company that didn’t have the staff change. Nobody invoices for it and everybody feels it.

How fast a replacement gets up to speedTwo ramps plotted over eight weeks. Without a shared record, a new lead's usable context climbs slowly and is still short of full at week eight. With a shared joint workspace, it climbs steeply within the first fortnight. The gap between the two curves is time the other company spends re-answering questions.how much of the joint context a replacement actually haseverything the last person knewday oneweek 2week 4week 6week 8reads the workspacehistory, decisions and open crossings, already writtenasks aroundrebuilds the project from other people's memoriesthe other company pays for this gap
The rampThe difference isn't the new person's competence. It's whether the joint history exists outside the person who left — and whether the company on the other side of the line has to reconstruct it out loud.

And there’s a second cost that’s harder to see. Trust doesn’t transfer. It was extended to a person, not to a logo, so the new lead starts from zero and spends a month proving they’re not going to break anything. Meanwhile the project’s real risk register — the things everyone had learned to watch — went out of the building with a laptop.

Continuity has to live somewhere neutral

The reason this problem persists isn’t that companies don’t value continuity. It’s that there has never been anywhere sensible to put the joint record. Put it in Atlas’s tools and Meridian loses it at the end of the engagement, or the moment someone in IT runs an access review. Put it in a person’s inbox and it leaves when they do.

None of this makes the two people less valuable. Judgment, relationships and the ability to read a room are exactly why they’re in the middle of it, and no system replaces that. What a system can do is stop the facts from being personal property — so that when someone hands over, they hand over their judgment and their contacts, and the six-month history is already sitting there, in one place, readable by both companies.

Two people should be the reason a joint project goes well. They shouldn’t be the reason it can’t survive a resignation.