Thirty seconds, down from ten minutes
Every modern Formula 1 pit wall runs machine-learning models that re-simulate the rest of the race every few laps: tire wear, pit windows, weather, what each rival is on and how long they can stay on it. When the question is whether to pit now or in three laps, strategy engineers describe getting the answer in roughly half a minute. A decade ago the same call could take the better part of ten minutes.
The role did not shrink with the calculation
Look at what stayed constant. Each team still has a named strategy engineer who makes the call, appears on camera after a good one, and answers for a bad one. The tool compressed the arithmetic. It left the accountability exactly where it was. Teams say that is deliberate.
Where the models stop
The models are strong on the closed part of the problem: how many laps a set of tires has left, how a stop reshuffles the running order, what a rival's pace implies about their plan. They have nothing on a driver mistake in the last five laps, a weather cell that has not yet appeared on radar, or a competitor about to do something unexpected. Those are weighed by a person with years of race weekends behind them, quickly enough to matter.
That boundary is why the structure around the engineer did not need redesigning. The tool answers a well-formed question quickly. Deciding which question matters this lap, and whether the answer can be trusted given what the model has not seen, remains a human skill, and the speedup gives the engineer more time for it.
A shape worth learning to see
Most conversations about AI at work are about a speed number or a headcount number. This case is neither: a decision that got roughly twenty times faster while the person making it stayed and, by the evidence, sharpened. The twenty-fold figure is a representative comparison for tire-strategy calls drawn from public interviews, not an average across every decision a team makes.
There are two wrong moves available. One is to hand the call to the model. The other is to leave the calculation with a person and a clipboard because that is how it has always been done. The pit wall does neither.
Your own work will have a few of these: a closed calculation with many inputs, a measurable result, a short window, and a person at the end who has to weigh what the numbers cannot see. Compress the calculation and spend the person's attention on the part the tool cannot reach. Most decisions will not fit. The ones that do are where AI earns its place quickly.
Finding those decisions is a short working exercise: Map the decisions in your team's work that look like this →
Keep reading: Williams Racing F1 and AI Transformation · Using it and deferring to it produce the same document, not the same person
This is the thinking behind a Trusted AI Culture.
Sources: Public Formula 1 team disclosures and technical-partner announcements on race-strategy modeling, 2022 to 2026. Cycle-time comparisons come from public interviews with strategy engineers; representative, not a single benchmark.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren