Chapter 5

The Friction Premium

Why Struggle Is the Method, Not the Obstacle

“The obstacle is the path.”

— Zen proverb

On the night of June 1, 2009, Air France Flight 447 was cruising at 35,000 feet over the Atlantic when ice crystals blocked its airspeed sensors and the autopilot, deprived of reliable data, disconnected. It handed a perfectly flyable aircraft back to its human pilots. What followed, reconstructed in the French accident bureau’s final report, took four minutes: a crew of professionals, confused by the very automation that normally protected them, held the plane in a stall it never needed to enter. Two hundred and twenty-eight people died in an aircraft with nothing mechanically wrong with it.

Aviation had seen this coming. Twelve years earlier, an American Airlines training captain named Warren Vanderburgh had given a now-famous lecture titled “Children of the Magenta Line,” warning that pilots were becoming systems operators who followed the automation’s magenta-colored path instead of flying the plane — and that their hand-flying skills were quietly eroding beneath them. After 447, the FAA made it official, issuing a formal safety alert urging airlines to have pilots practice manual flight, because the skills were measurably decaying from disuse.

Aviation is two decades ahead of the rest of us on exactly one question: what happens to human capability when the friction is removed. Every knowledge profession is now boarding that flight.

Desirable Difficulties

The deep irony is that learning science had the answer all along. The psychologist Robert Bjork spent a career documenting what he called desirable difficulties: conditions that make learning feel slower and harder while making it dramatically more durable. Henry Roediger and Jeffrey Karpicke showed that the effortful act of retrieving knowledge from memory — testing yourself, struggling to recall — beats comfortable re-reading for long-term retention, even though it feels worse while you do it. Generating your own answer before being shown the correct one strengthens learning more than being handed the answer first.

All of these share one structure: the difficulty is not an obstacle to the learning. It is the mechanism of the learning. Remove the friction and you remove the development — while the output, in the short term, looks better than ever. That is the trap in one sentence, and it is why the most dangerous quality of modern AI is not that it is sometimes wrong. It is that it is frictionless.

Struggle is not the obstacle to mastery. It is the mechanism of mastery.

The Sequence That Changes Everything

The single most practical principle in this book is not about whether to use AI. It is about when.

The professional who engages a problem first — even badly, even for twenty minutes — and then brings in AI is using the tool to extend thinking that exists. The professional who opens the tool first is using it to replace thinking that never happened. The outputs look identical. The people diverge, one shortcut at a time, along exactly the curve the pilots traced.

Depth in Practice: One Deliverable, Sequenced

The sequence rule sounds abstract until you lay it across a real week. The deliverable: a competitive analysis, due Friday, the kind of task AI appears born for.

Monday, thirty minutes, no tools: what do I already believe about this market, and what would change my mind? The output is a rough position — two competitors that matter, one that’s noise, a hunch about where the threat actually is. Tuesday, the AI gets the assignment and the position, with instructions to attack it: what am I missing, where is my hunch wrong, steelman the competitor I dismissed. The divergences — there are always divergences — become Wednesday’s work: the three load-bearing facts checked at their sources, one of which turns out to be a year stale, which is precisely the kind of discovery that never happens when the tool is trusted first. Thursday, the executive summary is drafted by hand before the AI polishes the supporting sections — the summary is where the judgment lives, and judgment is not delegated. Friday, the test that closes the loop: could you defend every claim in the room without the document? If yes, ship it.

Total additional human time versus the prompt-first version: perhaps ninety minutes across the week. Total difference in the professional: one of them investigated a market this week, and one of them read about a market this week. The client cannot tell the difference on Friday. They can tell by the third follow-up question.

The Practice

1. Identify the three tasks in your current work where the struggle itself is building your capability. Do them without AI for one month — deliberately, the way pilots now practice manual flight.

2. Before handing any task to AI, ask one question: is the value here in the output, or in the process of producing it? If the process — protect it.

3. Once a week, work one problem that exceeds your current ability, unassisted. The discomfort is not a malfunction. It is the workout.