I remember the exact moment I stopped believing the hype. It was 3:47 PM on a Tuesday, October 17th,

I remember the exact moment I stopped believing the hype. It was 3:47 PM on a Tuesday, October 17th, 2023. I was sitting in a 2023 Tesla Model 3, fresh off the lot from the Schaumburg delivery center, 37 miles on the odometer. I’d paid $49,990 for the car, then another $12,000 for Full Self-Driving Capability. Twelve thousand dollars. That’s more than a used Honda Civic. I wanted to believe. I needed to believe—not as a reporter, but as a guy who just blew a year’s worth of vacation money on a promise.

Here’s what nobody tells you about FSD. The name is a lie. It’s not Full Self-Driving. It’s a beta test you pay for. When Elon Musk says “beta,” he means it in the software developer sense—buggy, unfinished, prone to crashing at 2 AM on a rain-slicked Eisenhower Expressway off-ramp. But the marketing department kept the name because “Full Self-Driving” sells cars. I bought one.

I live in Chicago. That’s the first thing you need to understand. This isn’t Palo Alto, where the sun shines on perfectly painted lane lines and every intersection has a 5G tower. My commute runs from my apartment in Lincoln Park to the NBC Tower on the Magnificent Mile. It’s 7.4 miles. It takes 40 minutes in good traffic. FSD, in its current state, takes 55 minutes and three near-heart attacks.

The moment I realized this tech isn’t ready for normal people was on Lake Shore Drive, heading south past Belmont Harbor. It was a crisp November morning, 38 degrees, no precipitation. The car was in FSD Beta v11.4.7.3. I had my hands hovering, because you always hover. The car approached a construction zone with orange barrels narrowing the right lane. A human driver would slow down, check the mirror, merge smoothly. My Tesla did not. It accelerated. It aimed straight for a barrel like it was a target in a video game. I grabbed the wheel, yanked it left, and someone in a Ford F-150 honked at me. The car didn’t register the barrel at all. It was looking at the lane lines, not the obstacle. That’s the core problem: FSD sees paint better than it sees reality.

I’ve logged 847 miles with FSD engaged. I tracked every disengagement—every time I had to take over because the car was about to do something stupid. My count: 63. That’s one disengagement every 13.4 miles. In downtown Chicago, it’s every 3 miles. The system can’t handle unprotected left turns across multiple lanes of oncoming traffic. It hesitates like a teenager on their first driving test, then lurches forward when a CTA bus is 40 feet away. I’ve had it brake-check a pedestrian who was clearly waiting on the curb. I’ve had it try to turn into a one-way street the wrong way. Twice.

The price tag is where I get angry. FSD cost $12,000 when I bought it. Today, it’s $8,000. Tomorrow, Elon might raise it to $15,000. There’s no logic. It’s a lottery ticket with a 30% chance of working. Meanwhile, Ford’s BlueCruise is $800 a year. GM’s Super Cruise is $2,500 for three years. Both work on mapped highways, and both let you take your hands off the wheel. FSD doesn’t. You have to keep your hands on the wheel. So what did I pay for? The privilege of babysitting a robot that can’t merge onto the Kennedy Expressway without panic-braking.

Let me tell you about the Kennedy. It’s a 70-year-old concrete ribbon of hell that connects the suburbs to downtown. There are no clear lane lines in some spots. The pavement changes color every half-mile from road repairs. FSD hates it. Every time I take it, the car oscillates between lanes like it’s trying to decide if it wants to be a bus or a bicycle. I watched a YouTube video from a Tesla fanboy in California who claimed FSD handles “complex urban driving.” He lives in Irvine. Irvine has streets so straight you can see the curve of the earth. He’s never had to negotiate a left turn onto Western Avenue during a snow squall.

I did that last January. It was 19 degrees, snowing sideways, and the sensors were covered in slush within two minutes. The car threw up a “FSD degraded” warning and asked me to take over immediately. That’s fine. That’s honest. But what if I hadn’t been paying attention? What if I’d been checking my phone, like the marketing implies I can? The manual says you must “always be ready to take over.” But the name “Full Self-Driving” implies the opposite. That’s not a bug. That’s a feature designed to sell upgrades.

The worst part isn’t the failures. It’s the moments where it works perfectly. I had one drive last week, from my apartment to Andersonville for dinner, 6.2 miles. FSD handled every turn, every stop sign, every pedestrian. It was smooth. It was confident. I almost relaxed. I felt a little thrill—this is the future. Then, at a four-way stop on Clark Street, the car stopped, waited for a cyclist to pass, then lurched forward and nearly T-boned a Prius that had the right-of-way. The system had forgotten who arrived first. It reset its logic mid-intersection. That’s the kind of glitch that kills people.

I’ve interviewed three engineers who worked on autonomous driving at Waymo, Cruise, and Tesla. Off the record, they all said the same thing: Tesla’s approach—cameras only, no LiDAR—is fundamentally harder. LiDAR gives you a 3D map of the world in real time. Cameras give you a picture that a neural net has to interpret. If the net misinterprets a shadow as a child, you brake for no reason. If it misinterprets a child as a shadow, you don’t brake at all. Tesla has chosen the harder path because it’s cheaper. Your safety is the cost of that bet.

I’m not saying FSD is useless. It’s great on long highway drives. I used it from Chicago to Indianapolis in August, 180 miles, and it handled 95% of the driving. I didn’t touch the wheel for two hours. It was glorious. But that’s not revolutionary. That’s lane-keep assist with a fancy map. Nissan has that for $500. The promise of FSD is the city street, the unprotected left, the construction zone, the random guy walking his dog across an unmarked crosswalk at dusk. And FSD fails at those things consistently, predictably, and sometimes dangerously.

Would I buy it again? No. I’d buy the car—it’s a decent EV with good range. But I’d skip the $12,000 option. I’d use that money to buy a plane ticket to somewhere I actually want to go, because the car sure as hell isn’t driving me there. If a friend asked me today, “Should I get FSD?” I’d say rent a Tesla for a weekend. Drive it in your own city. Lie to yourself that it works. Then try to make a left turn during rush hour and tell me how you feel.

Here’s the reflection part, and I mean it: What I learned is that autonomy is not a software update. It’s a physics problem. It’s a human trust problem. We don’t need cars that can drive themselves 95% of the time. We need cars that can drive themselves 100% of the time, because the other 5% is where the accidents happen. Tesla is selling the 95% and asking you to cover the rest. That’s not Full Self-Driving. That’s a co-pilot who falls asleep. And you’re paying $12,000 for the privilege of keeping them awake.

I still own the Model 3. I still use FSD on the highway, sometimes. But every time I engage it, I remember that barrel on Lake Shore Drive. I remember my hands grabbing the wheel. I remember that, for one second, I trusted a machine to do what a 16-year-old with a learner’s permit can do. And it couldn’t. That’s the honest truth. That’s what nobody tells you. The future is coming. It’s just not here yet. And they’re charging you full price for a promise.