In June 2009, one of the safest aircraft ever built crashed into the Atlantic on the route from Rio to Paris. It was Air France Flight 447. A speed sensor froze and the autopilot did exactly what it was programmed to do: it handed control back to the humans. Experienced pilots, flying a functioning aircraft, could not react in time. The official report did not call them incompetent. What stayed from the case was something quieter: manual flying skill had eroded from simply not being used, and nobody noticed while it was happening.
I have carried that scene with me since I got to MIT. It explains the frustration of a long line of companies that invested heavily in artificial intelligence and are now asking why the financial return does not show up in the numbers. The lazy explanation is that the technology failed. What I see is a market making two structural management mistakes: buying today’s quick decision at the cost of tomorrow’s ability to decide, and confusing the creation of efficiency with the ability to capture margin.
A study at MIT’s Media Lab followed people writing with and without AI assistants. The assisted group showed lower brain connectivity, and many could not even recall what they had “written.” In another experiment, software engineers who used AI scored roughly 17 percentage points lower than the group working without it on a quiz about the very code they had just produced. AI does not make anyone less intelligent. It lets the unexercised skill atrophy. That is not an argument against AI. I use it every day and the gain is real.
For an operator, the expensive part of the technology is not the price of the subscription. It is the critical error your team fails to catch because it outsourced the exact judgment that would have caught it. Apparent productivity goes up while the ability to audit the machine goes down. When you outsource a core judgment, you are buying one answer and accumulating a debt of internal repertoire. The day the market shifts or the machine fails, the team will find out it no longer knows how to fly without the autopilot.
The second mistake is the illusion of value capture. There is a brutal difference between creating value and capturing it. Consider the yellow line on American football broadcasts, the virtual graphic marking the yard the offense has to reach. It is brilliant engineering, it changed the experience of watching the game, and the company that invented it captured very little of that value. The networks owned the relationship with the audience, and there were only a handful of them. The inventor did not set the price. His customer did.
The same leak happens with AI. If your company gains operational efficiency using a tool any competitor can also subscribe to, that efficiency runs straight through to the customer as a lower price. Your company produced more, but the gain did not stay with you. It was diluted across the market. Capturing value requires owning something exclusive: your data, a process that is genuinely yours, or a decision so sharp nobody can copy it.
The expensive mistake is inverting the logic: automating what makes you different and insisting on doing by hand what does not matter.
Before celebrating the productivity chart of your AI project, ask two questions at the board table. Are we buying this decision or building the ability to make the next ones? And what keeps the value we create today from leaking to the competition? If you cannot answer both, you are probably just funding a very pretty yellow line for the rest of the market.
