You have ten more years of driving, maybe twenty. After that it will be illegal.

Here is my prediction: before long, driving a car by hand will be against the law. Illegal as ordinary transportation, not as a hobby, and not everywhere on the same day. But illegal. And not to spite anyone. Because sooner or later, letting a tired, distracted, angry, or drunk human handle a ton and a half of metal at eighty miles an hour will strike people as exactly as insane as smoking on an airplane.

We smoked on airplanes for fifty years. We drove without seatbelts, we let kids stand up in the back seat, we got behind the wheel with blood alcohol levels that would cost you your license on the spot today. Every time, society eventually looked the practice in the face and thought: what on earth were we doing? Human driving is going on that list. It will not be the first technology tolerated out of habit and abolished out of clarity.

An anomaly, not an exception

Within fifteen years or so, a car crash will have become an anomaly. Not a rare exception. An anomaly. Something that happens because one of the two cars wasn’t autonomous.

The logic is simple. Today a crash is a statistical inevitability. We know there will be a certain number of them, we organize ourselves to bring that number down, and we accept the remainder as the price of mobility. In a fleet that is mostly autonomous, a crash changes nature. It stops being fate and becomes a bug. And a bug gets hunted, isolated, fixed, and pushed to every vehicle on the road overnight. No human driver has ever received an update that corrects, in all of his peers simultaneously, the mistake he just made.

People who fear autonomous cars are looking at today’s crashes. I’m looking at the curve.

The numbers, and what they can honestly be made to say

Tesla publishes its own safety statistics. The latest for North America: roughly 5.7 million miles between major collisions with supervised FSD engaged, against roughly 700,000 miles for the U.S. average. Seven to eight times fewer collisions. In Europe, where the system has only been on the road a short while, Tesla claims about four times fewer collisions than its own cars driven manually with the active driver aids switched on.

These are the manufacturer’s numbers, and I intend to treat them as such. Three caveats, stated plainly.

First, this is supervised FSD. A human is still behind the wheel, still liable, still ready to take over. It is not the Cybercab, which has no steering wheel and no pedals and whose track record still runs to thousands of rides rather than millions of miles. Second, the comparison suffers from everything a comparison can suffer from: route bias, with a share of highway miles probably above average; an in-house definition of “major collision”; and a benchmark that is the average American vehicle fleet, old and poorly maintained, not simply “the human.” Third, a paying robotaxi that runs empty, at night, through neighborhoods it did not choose does not yet have the same mileage behind it as a fleet of attentive owners watching every maneuver.

The one player that has been driving with nobody behind the wheel, at scale, for years now, is not Tesla. Waymo passed 170 million driverless miles at the end of 2025 and has since cleared 200 million, across Phoenix, San Francisco, Los Angeles, and Austin. Its data, published and cross-checked in several papers, all points the same way: fewer injury collisions, fewer intersection crashes, fewer pedestrians and cyclists struck. The same caution applies as with Tesla. Most of this remains Waymo measured against a local human benchmark, not an outside referee. Still, an argument about the death of human driving that cited only Tesla would be an argument about Tesla. Mine is an argument about transportation, and Waymo is currently its strongest evidence.

So no, the “ten times safer” figure you see everywhere is not a fact. It’s a slogan. And it weakens a case that stands perfectly well without it: several times safer, measured by several operators, with the curve still climbing. That direction is what cannot be argued with. Once a factor of four, five, or eight durably separates two ways of doing the same thing, the less safe of the two has no future. It has a deadline. And I have no desire to join the prophets behind the wheel of a Tesla.

What it looks like from the inside

I recently came across an account from a driver who has lived with these cars daily for five years, just back from two weeks in Florida with his family. Fort Myers, Naples, Miami, Key West. City streets, interstates, every condition. His verdict: stunning. Traffic circles, passing maneuvers, sudden slowdowns. No hesitation, no jerkiness. A precision and an efficiency he hadn’t expected.

He also described switching FSD on for a few minutes, out of curiosity, on Loop Road in the Everglades. Dirt track, no cell coverage, alligators in the ditch. The car followed the trail. Without flinching.

That’s an anecdote, not proof. A few minutes on a dirt road demonstrates nothing. But it illustrates the point that matters. An unmarked dirt track is the textbook case of “the robot will never handle that.” It more or less handles it already. And it is learning faster than we are.

Three leaps, not one

Look closely at the trajectory, because this is where most enthusiastic posts cheat. There isn’t one shift ahead of us, there are three, and they are not at the same stage.

The first is where we are now: very advanced driver assistance, with a human who remains legally and physically responsible. It’s what you can buy today, and it’s what France has just agreed to test.

The second is the driverless fleet. Waymo is there, in a handful of American cities. Tesla is barely entering it, with a few dozen geofenced Cybercabs inside one district of Austin. That is not yet national infrastructure. It’s a beachhead.

The third is the one that changes everything: vehicles that anticipate together. Today each autonomous car decides alone. If intent-sharing between vehicles becomes widespread, and nothing yet guarantees it at the scale of a country, one car will see another’s intention before it acts on it. The gap between two vehicles would stop being a human approximation, that blend of reflex, habit, and fatigue, and become a calculated parameter. Emergency braking would no longer exist as an event; it would be absorbed upstream by the entire line of traffic. And congestion, which is largely born of human reactions cascading backward, would dissolve on its own. No extra lanes to build. One human link to remove.

Three leaps, then. But all three pointing the same way, with no reversal observed anywhere.

The Cybercab, or the signal nobody wanted to read

On September 3, Tesla formally launched the Cybercab in Austin at an invitation-only event, with public rides opening immediately afterward within its Robotaxi fleet. Two seats, no steering wheel, no pedals. The first units came off the line at Gigafactory Texas in February, the ramp-up was announced in April, and the stated target for that line runs to hundreds of units per week. About fifty vehicles are currently registered in Texas. The federal regulator, NHTSA, promptly opened an inquiry into the vehicle’s certification, which Tesla self-declared compliant with federal safety standards.

Fifty cars in one district of Austin is nothing. And that is precisely what I’m asking you to look at differently. Elon Musk is not building a car that drives itself. He is building transportation infrastructure. Those are not the same thing. A car is something you buy, compare, replace. Infrastructure gets installed city by city, spreads, and one day becomes the standard against which everything else is judged obsolete. Waymo followed exactly that path: one city, then four, then roughly twenty announced for 2026, London and Tokyo among them. The Cybercab is not another model in the catalog. It’s the building block of a network, and a block designed from the outset never to accommodate a driver.

And France in all this? France is negotiating

You might have expected Europe to hold out longer. In July, the Ministry of Transport judged FSD too risky for French roads. Six weeks later, following a “constructive exchange” with Elon Musk, minister Philippe Tabarot announced that two Teslas running supervised FSD would be driving on French roads as part of a trial, and that the authorities had been working with the manufacturer for months on the technical adaptations needed to support European type approval, due for a vote in early October. Five European countries had already taken the step. Tesla, for its part, was careful to point out that trials have been running in France for two years.

I’m not going to claim France capitulated. Two cars at the UTAC test center and a committee vote is not a surrender; it’s industrial diplomacy ahead of a deadline. But look at the mechanism. A refusal on principle, a video call with the CEO, a reversal in six weeks. It isn’t a scandal, it’s an admission. The government knows that in this race, standing still costs more than taking the risk, and a European auto industry already in trouble cannot afford to miss autonomy the way it missed other shifts. It’s falling into place gradually, quietly, and barely disguised. Two test cars today, type approval in a matter of weeks, robotaxi fleets in our cities within a few years.

It won’t start with a law, it will start with a premium

Nobody is going to outlaw human driving by decree, at least not at first. It will happen through insurance, the way these things always do.

The day an actuary can lay two claim rates side by side, measured across millions of miles and several fleets, with one several times lower than the other, he will do his job. The human driver will be classified as unreliable, like a smoker on a life policy or a young driver in a sports car. His premium will climb. Then it will double. Then insurers will start declining the risk outright, the way they decline certain repeat offenders today. At that point no ban is necessary. Nobody will want to drive anymore, for the simple reason that nobody will be able to afford it. The law will merely ratify, years later, a reality the market will already have settled.

And when that law arrives, it will surprise no one. It will be received the way the smoking ban in restaurants was received: with a touch of nostalgia for a few people and relief for everyone else.

The analogies are not hard to find

We got rid of elevator operators once the elevator could pick its own floor. We unplugged the switchboard operators once the exchange could route calls without them. We never allowed horses on the highway, not out of contempt for the horse, but because an unpredictable animal in the middle of a predictable flow endangers everyone. We banned lead in gasoline, asbestos in walls, and train drivers without automatic speed control.

Every time, the same story. A practice tolerated because we didn’t know how to do better, then abolished the day we did. Human driving will not be the exception. It’s next on the list.

What could prove me wrong

The timeline, first. France recorded 3,263 road deaths in mainland France in 2025, and 3,515 including its overseas territories. The magnitude is appalling, but a good share of those deaths are not “one car into another at eighty.” They are motorcycles, pedestrians, single-vehicle departures from a rural road at night. Autonomy does not erase that overnight, and certainly not everywhere at once. The ban will arrive in layers, and I put the first one ten years out: the major arteries, where traffic is most predictable and the gain most measurable. Metropolitan areas will follow close behind. The rest, the snowbound mountain road, the construction site, the tractor, the motorcycle, will hold out longer, perhaps a generation. But once driving yourself between two cities becomes impossible, the outcome is already decided. What survives is no longer transportation; it’s a residual use, tolerated for the time being.

The players, second. Nothing says it will be Tesla. The turning point may well be named Waymo, or a Chinese manufacturer whose name we haven’t learned yet, and Europe may well end up on the receiving end rather than in the driver’s seat. None of that changes the prediction. It only changes who absorbs the blow.

What it means for us

I don’t welcome all of it. There is a measure of freedom in holding a steering wheel, and a measure of responsibility too, and I know exactly what it means to reach that moment when the driver, the machine, and the road stop being three separate things. I can already picture the private circuits, the closed roads, and the clubs where people will go one Sunday a month to drive themselves, the way people go horseback riding today: for pleasure, within a frame, and at their own expense. It will be a pastime. It will no longer be transportation.

But I’m not going to pretend the curve will reverse. It won’t. You have ten more years of driving, maybe twenty. Make them count.

Hallelujah

One last thing: writing this piece has left me with a violent urge to drive. Not to be driven. To drive. To feel the thing my own prediction is going to make illegal.

So I’ll stop here. My car is out of the garage, the engine is warming up, and it’s waiting for me.

Hallelujah. 🙏🏻