A study dropped this month showing what everyone feared. Give someone five minutes with an AI assistant and their ability to persist through hard cognitive work measurably collapses.
The effect is immediate and sizable. Hand someone a problem they could solve alone, let them watch an AI solve it instead. When you take the AI away they fold faster on the next task.
This exact panic arrived fifty years earlier, just with arithmetic instead of reasoning. When calculators entered American classrooms in the 1970s, controlled studies documented precisely what you'd expect. Students who used them showed measurable drops in mental arithmetic fluency.
By 1980, the evidence looked conclusive — calculators were cognitive crutches, the research was sound, the fear justified. Within a decade the prediction had inverted. Math achievement didn't collapse. It rose. Not because students suddenly developed superhuman computational persistence. Because math teachers stopped teaching arithmetic as the goal and started using freed cognitive capacity for problem-solving that actually mattered.
What matters is not whether AI offloading narrows your capacity for computation by proxy, but whether you're spending the freed persistence on work that demands it more.
”The AI study measures something real — the persistence drop is not fictional. But what it measures is whether you can solve a problem after watching someone else solve it, not whether you've built a better tool for thinking. A writer using AI to draft boilerplate isn't losing her mind. She's lost if that's all she does with the time. She's found it if that hour goes toward the sentences that only she can write, the ones that require something no language model can generate. Which is precisely the moment when persistence stops being a generic cognitive trait and becomes the shape of actual thinking.