I Tested Tesla’s Full Self-Driving System In New York City. It Was Embarrassing

Last week, I drove the 2026 Tesla Model Y in Brooklyn with its Full-Self Driving (FSD) system engaged. My hands hovered over the steering wheel, but the car handled the braking, acceleration and turning in the worst place in America to drive.

I was on a two-way street, with a cargo truck unloading right in the middle of the road, parked just a few feet away from a traffic light ahead. Horns blared, delivery riders snaked around and there was a chain of vehicles approaching from the opposite direction. It would be a stressful situation even for a skilled human driver, but there I was, trusting an automated driver assistance system to navigate the chaos.

Initially, the Model Y handled the mess like a pro. It slowed down, waited for a gap in oncoming traffic, then neatly swung around the cargo truck in a reassuringly human-driver sort of way. I was impressed, at least for a moment.

Photo by: Suvrat Kothari

But as soon as it stopped at the red light, a new problem was already brewing. Another truck aggressively swung into the intersection for a wide right turn. Its sheer length and long turning radius pushed it straight into my lane. The Model Y just sat there, waiting for the light to turn green, while the truck kept coming.

At the last second, I hit the accelerator and nudged forward to avoid what would have been a very costly brush between the corner of the truck’s bumper and my driver’s door.

That’s the thing about driving in New York City. You clear one obstacle, and the next one is already unfolding in front of you. It’s constant, unpredictable, relentless and generally not advisable. It’s the kind of environment that tests the limits of any driver, human or robotic.

Photo by: Suvrat Kothari

Tesla’s approach to “solving” autonomy involves relying on cameras and AI. The company uses multiple cameras acting as the vehicle’s “eyes.” This visual data feeds into machine-learning models, which then make sense of it and help the car make active decisions based on what it “sees.” Autopilot is Tesla’s highway driving assistance system, and FSD takes things further with in-city and in-town navigation, lane changes, turns at intersections, stops at traffic signals and more.

CEO Elon Musk argues that this vision-based approach is cost-effective and easily scalable, as all Teslas are already equipped with cameras. But experts have long warned that this approach could be ineffective.

That’s about to matter in New York more than ever. The FSD system on this Model Y is essentially the same software Tesla is loading into its Robotaxis, which are also Model Ys with decals and an Uber-like ride-hailing app. As Tesla and Waymo plan to expand operations across the U.S., New York City is primed to be their ultimate test. It’s the country’s most densely populated city. Its traffic landscape means countless edge cases—unpredictable situations that can confuse autonomous systems, like a truck barreling towards you or a gridlock in the middle of a street.

Photo by: Suvrat Kothari

If Tesla can make Robotaxis cars work here, it can likely make them work anywhere else. But so far, even in places like Austin and the Bay Area, where Tesla has already started its ride-hailing service, true autonomy is still out of reach. New York State laws still require Tesla to have test drivers behind the wheel, ready to step in. Still, last year, New York’s Department of Motor Vehicles opened the door to autonomous vehicle testing, granting the first permit in August to Waymo, Alphabet’s robotaxi arm, for testing in Manhattan and Brooklyn with a human test driver behind the wheel.

Around the same time, Tesla began hiring robotaxi test drivers in New York, InsideEVs previously reported. (The job listing has since disappeared from Tesla’s careers site, though the company is still hiring operators in California, Texas and Florida.) I also spotted Model Y Robotaxis testing in Brooklyn, but with human drivers behind the wheel. As of August, Tesla had yet to apply for an official testing license in the city, a spokesperson for the NYC Department of Transportation told CNBC.

Waymo and Zoox take a very different approach than Tesla does. Those cars are loaded with a full sensor suite, including cameras, radar and lidar. Lidar is expensive, but its prices are dropping quickly. Elon Musk has long rejected it, calling lidar a “crutch” and saying any company that relies on it is “doomed.” Meanwhile, Waymo is proving him wrong in practice. The company now operates in multiple U.S. cities and handles more than 250,000 truly driverless rides every week.

That said, the run-in with the turning truck wasn’t the only time I had to intervene during the three-odd hours I spent in the city testing the system. The Model Y didn’t stop or even slow down when a school bus retracted its stop signs. When an FDNY truck came across roaring, with its sirens blasting and lights flashing, the Tesla didn’t yield.

I’ll also never forget the look on a cyclist’s face when the Model Y confidently swung left across a bike lane in Lower Manhattan, forcing him to brake—I did not intervene because there was still a decent amount of room for me to pass, but it still wasn’t a courteous maneuver. FSD, even in Chill mode, tends to be pretty assertive.

In Standard and Hurry modes, FSD is borderline overconfident. At one point, it even drifted over the yellow line in Jersey City, eerily similar to how this Robotaxi behaved during the pilot program in Austin.

Tesla Model Y with FSD drove past a school bus in Brooklyn with its stop signs retracted without slowing down.

Photo by: Suvrat Kothari

Still, there were plenty of instances where FSD impressed me. But those instances were all out of the city, in the suburbs, where traffic is less chaotic. When I left New York and drove down to visit friends in the suburbs of Washington D.C., FSD was effortless. On a clear, sunny day, FSD did the vast majority of the highway drive with me supervising, making the 230-mile journey feel much shorter than it actually was. The next day, it carried us 15 miles to a restaurant and back without a single intervention.

Later that evening, when the sun went down, the Model Y’s excellent matrix headlamps lit the road so brightly that the cameras seemed to have no trouble seeing ahead. Its navigation skills stood out, too. You shouldn’t miss an exit or on-ramp ever with FSD engaged.

If marketed correctly, FSD is still one of the best Level 2 advanced driver assistance systems out there. Setting aside its misleading name, it does more than other Level 2 systems like General Motors Super Cruise or Ford’s BlueCruise. The difference is that those systems do exactly as advertised, whereas FSD overpromises and underdelivers. It’s far from what the Full-Self Driving label suggests and Tesla itself recommends complete driver supervision, including hands on the wheel and readiness to take over at all times.

Now, as Tesla pivots away from being just an EV maker and rebrands itself as an AI and robotics company, the company would need to demonstrate that FSD can solve complex, edge-case traffic situations consistently and reliably. Its ability to do that could be the difference between Elon Musk getting paid a trillion dollars or getting paid nothing. Until FSD can handle the chaos of New York without constant human backup, his dream of a truly driverless future remains more of an experiment than reality.

Have a tip? Contact the author: suvrat.kothari@insideevs.com