Who's Actually Running Your AI?
There’s a soccer ball at the 2026 World Cup that someone has to charge.
Let me back up. This World Cup might be the most automated sporting event ever staged. The official ball has a sensor inside it that reports its own position around five hundred times a second. Every one of the 1,248 players sat for a quick 3D body scan, so the system can track a shoulder or a knee down to the centimeter. When a player’s clearly offside, a beep goes straight into the referee’s earpiece, before the linesman even lifts the flag. Sixteen cameras per stadium, more than a hundred million tracking points per match. It looks like the machines finally took the whistle.
And yes, the ball has a battery. Somebody on the equipment staff plugs it in before kickoff (a ninety-minute charge gets you about six hours of play). The most advanced officiating tech ever built can be undone by a dead battery, which is the kind of detail I could think about all day.
But the battery isn’t the surprising part.
The night shift behind the broadcast
All of that “AI” runs on people. Around the world, in cities like Manila, Cairo, Chennai, and Rio, data annotators watch the same matches you do and log every pass, every tackle, every shot, by hand. One report found a single annotator can tag up to three thousand actions in one game, spend three or four hours doing it, and earn roughly seventy dollars. That hand-labeled work is what teaches the cameras and the smart ball what they’re even looking at.
A researcher who studies these workers said it better than I can: the football we watch runs on their work as much as on the players.
The spectacle gets sold as autonomous. Underneath, it’s a global night shift of humans, tagging plays for about the price of a nice dinner, so the system on TV can look like it’s thinking. What’s true at the World Cup is true of almost every AI system you’ve ever been sold.
The demo is the tip
When a vendor shows you a slick AI demo, you’re seeing the polished ten percent. Underneath sits the ninety percent nobody demos: the data labeling, the human review, the exception handling, the people fixing what the model gets wrong before you ever see it. The model is what gets the press release, but the people are what make it work.
I’ve watched companies buy technology for twenty-five years (AI for a few of those, software for all of them), and the pattern repeats. Leadership budgets for the model and forgets the humans. Then the thing ships, and someone discovers that “automated” actually means “automated, plus a team in another time zone catching the edge cases.” The cost moved somewhere the org chart doesn’t show.
So if you’re rolling out AI this year, here are three things the World Cup will teach you for free.
One: budget for the humans you can’t see
The headline number on any AI project is the model. The real number includes the people who label the data, the people who check the output, and the people who handle whatever the model punts on. Skip that math and the project still ships. It just costs more than you said it would, and the savings you promised the board never quite show up.
Two: automate the measurement, not the judgment
Notice what FIFA actually automated. The system measures position to the centimeter. A human referee still decides whether an offside player interfered with the play. They built it that way on purpose. The machine is great at “where was the knee” and terrible at “did that matter.” The companies getting AI right draw the same line: let the model measure, count, draft, and flag, and keep a person on the call that needs judgment, taste, or accountability. The phrase “semi-automated” is doing a lot of honest work.
Three: when everyone has the same tool, the question is the edge
FIFA gave the exact same AI tactical tool to all 48 teams, with the same data and access for everyone. So the advantage comes down to what you ask it.
That’s true in your business too. The model your competitor uses is probably the model you use. Once the tool is a commodity, the edge moves to whoever asks the sharper question of it. Which, you’ll forgive me for pointing out, has been my whole thing for a while.
The humans never left
AI gets sold as a way to remove people. Mostly, it just moves them: from the front office to a back room, from the visible part of the process to the invisible part, and too often from a salary to a piece rate. The humans never left the World Cup. We just stopped pointing the camera at them.
You can take it seriously without feeling guilty. Be honest that it’s happening, in your company and everyone else’s. The most useful thing you can do with AI this year is to ask who’s really doing the work, what they’re being paid, and which decisions you should never hand off, no matter how good the demo looks.
If the most advanced AI on the planet still runs on people you’ll never see, it’s worth asking what’s running yours.