In 2002, the U.S. military staged the largest, most expensive war game in its history — a simulation of a future Middle East conflict, run by a team of Pentagon planners playing the American side.
Their opponent was a retired Marine general named Paul Van Riper, brought in specifically to play the enemy commander and test whether all that planning could survive contact with someone who refused to cooperate.
The planners had a move ready. Cut his microwave towers, cut his fiber-optic lines, and he’d have no choice but to fall back on satellite phones and cell towers — channels they could intercept and track.
It was a clean, mechanical move. Remove part X, force a predictable response Y. The kind of thinking that works beautifully on an actual machine.
Van Riper wasn’t running a machine. He switched to motorcycle couriers. He hid written orders inside daily prayers. He signaled his aircraft with a World War II lighting system that no modern sensor was built to catch.
The planners’ mistake wasn’t really about having too much data, though they had plenty — forty thousand database entries, a framework sorting the enemy into tidy categories. The bigger mistake was treating an adaptive opponent like a complicated system: predictable, decomposable, responsive to the right lever pulled in the right place. Cut the comms, force the response, win the exchange.
Complex systems don’t work that way. There’s no fixed relationship between cause and effect when the thing on the other end is capable of genuinely inventing a new move nobody accounted for. Van Riper sank sixteen ships in an afternoon, twenty thousand simulated casualties, because he was never playing the game the planners’ model assumed he’d play.
I think about this every time a company announces layoffs and calls it “AI restructuring.”
The underlying logic is the same mechanical substitution. Swap AI in, subtract headcount out, and the spreadsheet balances. It’s the language of a complicated system — remove this part, install that one, calculate the net effect.
Except a workforce isn’t an engine. Client relationships don’t transfer cleanly. Nobody backs up institutional knowledge and restores it intact. Morale among the people who stay doesn’t move in a predictable, calculable direction.
A Gartner study of 350 firms this year found that the companies cutting hardest showed no actual improvement in returns — suggesting the mechanical model quietly failed the same way the war games did. The swap looked clean on paper. The system it was swapped into never actually behaved like a machine.
“AI restructuring” sounds like an engineering decision. It’s usually a complex, human transformation wearing the language of a complicated one, because complicated problems sound solvable and complex ones sound like something you’d have to actually sit with.
The Pentagon’s planners lost because they assumed the enemy would respond like a machine. Many companies right now are betting their workforce will absorb a machine-sized cut without behaving like the complex, adaptive system it actually is.