The $125 Million Coordination Bug

September 23, 1999. NASA's Pasadena lab.

At around 9:00 AM, a small group of engineers tracked the Mars Climate Orbiter executing its final maneuver. The $125 million spacecraft was approaching Mars after nine months in space. Routine. No drama expected.

Then the signal went dead. The screen stayed black. The orbiter never came back.

This wasn't the kind of failure you'd expect from NASA. They had simulated every contingency. The spacecraft was small but well-built. The mission profile was conservative. Two organizations had worked it: Lockheed Martin, the aerospace contractor that built the spacecraft and calculated its thruster forces. And NASA's Jet Propulsion Laboratory, which navigated it through space.

Both teams were the best in the world at what they did. The error wasn't in either team's work.

Lockheed had done the thruster math in pound-force. The JPL navigators had assumed it was in Newtons. The numbers were off by a factor of 4.45×.

This is just unit conversion — the kind of thing a freshman engineering student would catch. But because pound-force and Newtons sit in the same numerical ballpark, nothing seemed out of place.

Each team trusted the other's competence. These were experts working with experts. Why double-check?

Worth pausing on. Both teams were excellent at their job. Both teams trusted the other. The thing that failed was the interface between them — the place where Lockheed's outputs became JPL's inputs.

This is where every coordination failure lives. Not inside any team's competence. Between them.

And in 1999, that interface was just a number transmitted from one document to another. No system caught the unit mismatch. No system was watching.

Over nine months and 49 million miles, every thruster firing drifted the spacecraft slightly off course. Each one too small to notice. Each one compounding.

By the time the orbiter neared Mars, the navigation team thought it was 226 kilometers above the surface — a safe orbital insertion altitude. It was actually 57 kilometers above the surface — inside the atmosphere.

The thin Martian air tore the spacecraft apart in seconds.

Now scale this from one spacecraft to a Fortune 500. Replace Lockheed and JPL with your sales team and your engineering team. Replace pound-force vs Newtons with their schema definitions, their KPI definitions, their tools, their assumptions about the customer, their assumptions about "ready to ship."

Every enterprise has a dozen Mars Climate Orbiters in flight at any moment. Two excellent teams, working in different units, trusting each other.

You usually find out years later. Sometimes you never find out — you just know the company underperforms its talent.

Talent solves complexity within teams. Coordination solves the complexity between them. When coordination fails, no amount of expertise can bridge the gap.

— From Reshuffle, Chapter 6

The most underrated thing AI does in the enterprise isn't automate any individual job. It's watch the interfaces between teams — translate one team's vocabulary into another's, flag when two outputs should have agreed and don't, build the shared representation NASA couldn't.

The $125M lesson is now affordable everywhere.

Sangeet on this in Chapter 6 ↗