Every Failure Lives at a Seam

When something goes badly wrong in a company, the first instinct is to find the part that broke. Which system failed? Which team dropped it? Whose fault was this? After four decades of building companies in fintech, crypto, blockchain and now AI, I have come to distrust that instinct. Most of the worst failures I have seen were not inside any one system or any one team. Each side worked. Each side could prove it worked. The failure lived in the handoff between them.

I call those handoffs seams, and I have learned to look there first.

The money is in the seam

Payments makes this lesson unusually concrete, because in payments a seam has a dollar amount attached. One system records that the money left. Another system records that it never arrived. Both logs are accurate. Both teams can show you their records, and the records are right. And the money is somewhere in between, in the gap that each side assumed the other was watching.

The first time I saw this happen at scale, the hardest part was not finding the money. It was finding anyone whose job it was to look. Every person in the room had a clear responsibility, and every responsibility ended at the edge of their own system. The space between two correct systems was nobody's.

That pattern does not stay in payments. Once I learned to see it, I saw it everywhere.

Where seams hide

A salesperson promises a customer something that the product team has never heard of. Sales did its job: it closed. Product did its job: it built what was planned. The customer falls into the gap between them, and nobody notices until the renewal.

Two services exchange data, and one writes a date in one format while the other reads it in a different one. Each passes its own tests. Together they are wrong on certain days of every month, and the bug takes weeks to find because both teams, with good reason, are sure their side is fine.

A new customer is handed from onboarding to support and has to explain their whole situation again, because the notes did not travel. No one did anything wrong. The context simply had no owner once it left the first team's hands.

A vendor makes a small change. An integration that was set up years ago, by someone who has since left, and that nobody ever wrote down, breaks quietly. The vendor did nothing improper and your system did nothing improper. The connection between them belonged to a person who is no longer there.

Why seams get missed

Seams are invisible for a structural reason. Organizations are drawn as boxes: teams, departments, services, vendors. Accountability is assigned to boxes. Metrics are reported by boxes. Every incentive in a company tells each box to make itself correct, and almost nothing tells anyone to make the space between boxes correct.

Specialization makes it worse as a company grows. The more expert each team becomes in its own part, the less any of them knows about the edges of the others. Early on, a few people hold the whole thing in their heads and the seams take care of themselves. Later, the seams multiply faster than anyone notices, and the people who used to cover them are busy running their own departments.

There is also a human reason. Owning a box is rewarding: you can point to it, improve it, defend it in a review. Owning a seam is thankless. When it works, nobody notices, and when it fails, the owner is the first person in the room to be asked what went wrong. So unless someone deliberately assigns it, nobody volunteers for it, and the most important ground in the company stays unclaimed.

AI is now adding seams of its own. Every time a model hands work to a tool, or one agent passes a task to another, there is a new handoff, often with less context than a person would carry across it. The same rule applies there.

What I do now

I do four things, and none of them is clever.

First, I draw the seams, not just the boxes. When I look at an architecture diagram or an org chart, I ask for a second picture: every point where one part hands something to another. That picture is usually messier than the first one, and more useful.

Second, every seam gets a named owner. A person, not a team. Teams cannot own a gap; they will each assume the other one does. A person can be asked, on any given day, whether the handoff worked.

Third, I reconcile across seams instead of asking each side to report on itself. It is not enough for each team to say it did its part. Someone has to check that what left one side actually arrived on the other, in the form it was supposed to, every day. Payments people call this reconciliation. Every business needs it, whether or not money is involved.

Fourth, when something breaks, I look at the handoff before I look at either side. It saves days, and it saves a great deal of unfair blame, because the people on each side were usually doing their jobs well.

The gaps everyone assumed were covered

The most dangerous sentence in a company is not "that is not my job." It is "I assumed they had it." People who say the first are at least telling you where their responsibility ends. People who say the second, with complete sincerity, are standing at a seam nobody owns.

Companies rarely fail where everyone is looking. They fail in the gaps that each side assumed the other side was watching. Find those gaps. Put a name on each one. It is unglamorous work, and it is some of the most valuable work a founder can do.