Going live in 3 months
Industry

Before ChatGPT, There Was Autopilot

Elias Mwewa · · 6 min

On January 17 this year, I got into a Tesla for the first time, and it was genuinely exciting. Thanks, Benjamin Wright.

0:00

The cabin was quiet in a way gas cars never are. A giant touchscreen did the job of most of the buttons and dials. Sitting there, I kept coming back to one thought: this isn't a car with some software in it. It's a computer on wheels. I was excited and asking my friend some questions. The video cut short but not the feeling of that ride.

That ride sent me down a rabbit hole, and I came out with a take I don't hear often enough. Tesla was one of the first companies to put artificial intelligence to real, practical use, years before ChatGPT made AI a household word. Yet today, almost all of the praise goes to OpenAI and Anthropic.

Those labs have earned their credit. But Tesla has been quietly succeeding in AI for a long time, and what it's attempting right now is one of the boldest real-world AI bets in the industry.

Tesla was doing AI before it was cool

Tesla's AI story starts in October 2015, seven years before ChatGPT launched. That month, a software update switched on Autopilot, letting cars steer, change lanes on the highway and parallel park. Tesla said about 60,000 cars already on the road had the hardware for it.

The bigger idea was the fleet. Even then, Tesla said Autopilot would keep improving as owners drove, using data collected from its cars. That is the same principle behind today's AI boom: more real-world data makes a better model.

In 2019, Tesla went further and started designing the computer itself. At its Autonomy Day event that April, it revealed its cars had already switched from Nvidia processors to an in-house FSD chip built to run neural networks. Tesla was building in-house what most automakers still buy from suppliers.

Then, in January 2024, Tesla began rolling out FSD v12. It replaced more than 300,000 lines of hand-written code for city driving with a single neural network trained on millions of video clips. In plain terms, the car stopped following rules engineers wrote and started learning to drive from examples, the same basic recipe behind a chatbot.

Why the chatbots get the spotlight

Chatbots get the credit because you can talk to them. Anyone can open ChatGPT or Claude and see the intelligence in a single reply. When Tesla's AI works, it's invisible: the car changes lanes smoothly and you barely notice.

Tesla also gets filed under "car company." Conversations about it drift to delivery numbers and the stock price, not the neural networks underneath.

That's a blind spot, because the two worlds overlap. Andrej Karpathy was a researcher at OpenAI before spending five years leading Tesla's Autopilot vision team. Same talent, same core techniques, very different stakes.

To be fair, the labs earned their moment. Language models can write, code and analyze across almost any subject, and that breadth is a real breakthrough. But when a chatbot slips, you get a bad answer; when a car slips at highway speed, the consequences are physical.

What Tesla is chasing right now

Tesla's current bet is simple to describe and hard to pull off. It wants its driving AI to run with nobody at the wheel, and to build a business on that.

The clearest sign is Robotaxi. As of July, Tesla's ride-hailing service was live in seven U.S. metros. In six of them (Austin, Dallas, Houston, Miami, Orlando and Tampa), it was ramping up rides with no safety driver on board.

This month, Tesla added the Cybercab, a two-seater with no steering wheel or pedals. Public rides began in a limited part of Austin on September 4, a day after an invite-only launch, with 45 Cybercabs registered at the time. Tesla has also opened an interest form for businesses that want to buy Cybercab fleets.

FSD itself is becoming a subscription business: Tesla reported 1.48 million paying customers at the end of June, up 56% in a year. More than 55% of new North American deliveries included it, and the Netherlands, Belgium, Denmark, Estonia and Lithuania have approved the supervised version.

Then there's Optimus, the humanoid robot. Tesla retired the Fremont lines that built the Model S and Model X and is installing Optimus production lines in their place, with output expected later this year. The first robots will collect training data, the same data-first playbook Tesla used with its cars.

Underneath it all, Tesla keeps pulling the AI stack in-house. It more than doubled its AI training compute in Texas in the first half of 2026, and it has started building its own chip fab in Austin.

My read: Tesla wants to own every layer of real-world AI, from the data and chips to the cars and robots that use them. If that works, it stops being judged as a carmaker and starts being judged as an AI company with a fleet.

What critics get right

Tesla isn't the leader in driverless rides yet; Waymo is. On the day Cybercab launched, Waymo had about 4,000 autonomous vehicles giving paid rides in 14 U.S. cities. Even in Texas, Tesla's home turf, Waymo's fleet was more than twice the size of Tesla's.

The FSD in customer cars is still supervised. Tesla's own quarterly update notes that it requires active driver supervision and does not make the car autonomous.

The timelines have slipped for a decade. In 2015, Elon Musk said full autonomy might be about three years away. A January 2025 prediction of roughly 10,000 Optimus robots built that year was also missed.

Regulators are watching, too. NHTSA is evaluating the Cybercab rollout and separately probing how Tesla's camera-only system handles fog, glare and dust. That probe involves at least nine crashes, including one that killed a pedestrian.

And camera-only is still a live debate: Waymo has said cameras alone aren't enough. None of this erases Tesla's AI record. It just means the most ambitious part of the story hasn't been proven yet.

Why I want one

That January ride won't be my last. I want to own a Tesla. I want to know what it feels like to ride in one to work every day, and to every place I visit across the United States.

To me, it's one of the best use cases of artificial intelligence out there. Chatbots are AI you talk to. A Tesla is AI you ride in.

The AI labs deserve their headlines. But when we list the companies that made AI real for everyday people, Tesla belongs near the top, and it got there years before most of us were paying attention.

Sources

Comments 0