Nobody's Behind the Wheel

It's Monday, July sixth. I'm pretty sure I saw a Tesla robotaxi driving around Santa Clara last week. It went by pretty quickly so I couldn't see if there was a human operating the vehicle, but I can tell you it looked pretty slick. All gold, covered top to bottom. Looked very futuristic. And then I was at dinner with a friend over the weekend who owns a Tesla with Full Self-Driving, and I asked him about it. He said he can't be on his phone or totally zone out while in the driver's seat because the car still needs him to intervene regularly, and that he's never had a trip where FSD handled everything without him stepping in at some point. Which raised an obvious question: if the consumer version of Tesla's self-driving software still needs a human watching it constantly, what's different about the robotaxis that Tesla is now running in five American cities, some of which are operating with no human in the car at all? Is it a different system? Is it just the same software with the safety driver removed? And how does it compare to Waymo, which has been doing this for years with a completely different approach to how the car sees the road? I spent the weekend looking into it, and the picture is more complicated and more interesting than I expected. Tesla launched its robotaxi service in Austin, Texas in June 2025 with ten cars and a human safety monitor sitting in the passenger seat. In January 2026, it started running some vehicles without the safety monitor. By April, it expanded to Dallas and Houston. Last week it launched in Miami, making it five cities, and Miami is the first deployment outside of Texas.

The scale is worth understanding because the headlines make it sound bigger than it is. As of recent Texas DMV filings, Tesla had about 42 active robotaxi vehicles in the state. Waymo, by comparison, had 577 registered driverless vehicles in Texas and was delivering over 250,000 trips per week across all its markets. Tesla's robotaxi is a pilot program. Waymo is a commercial operation. The gap between them is enormous.

The early results for Tesla's service have been mixed. NHTSA, the federal auto safety agency, opened an investigation after videos surfaced of Tesla robotaxis driving on the wrong side of the road, phantom braking in intersections, dropping passengers off in the middle of streets, and committing traffic violations. Tesla uses the same Full Self-Driving software that powers the consumer version in people's personal Teslas, just running in a mode that doesn't require a human to be paying attention. The software is the same. The difference is that the safety net has been removed. Your friend's Tesla still makes mistakes but he catches them. The robotaxi makes the same mistakes and nobody catches them. The fundamental disagreement between Tesla and Waymo isn't about business strategy or pricing. It's about how a car should see the world, and whoever is wrong is going to lose billions of dollars finding out.

Tesla uses eight cameras and nothing else. The car sees the world the way you do: through lenses that capture two-dimensional images and then uses AI to interpret what those images mean in three-dimensional space. Elon Musk has called LiDAR, the laser-based sensor that creates three-dimensional maps of the environment, "a fool's errand" and "a crutch." His argument is that humans drive with eyes, which are essentially biological cameras, and that if you build good enough software you don't need anything more than cameras to drive safely.

Waymo uses 29 cameras, five LiDAR units, and six radar sensors per vehicle. Their philosophy is that redundancy saves lives. LiDAR generates centimeter-accurate three-dimensional point clouds that work in all lighting conditions, including darkness, rain, and glare where cameras degrade. Radar measures the speed and distance of objects even when visibility is poor. When sensors disagree with each other, the system fuses all available data rather than relying on any single input. The argument is that driving is too safety-critical to depend on one type of sensor, no matter how good your software is.

This isn't an academic debate. In March 2026, federal regulators opened an engineering analysis, which is the final step before a potential recall, after finding that Tesla's camera-only FSD "fails to detect and/or warn the driver appropriately under degraded visibility conditions such as glare and airborne obscurants." Miami's launch is particularly significant because South Florida has frequent intense rainstorms and harsh sun glare, which are exactly the conditions where cameras perform worst and where LiDAR's advantages are most pronounced. If Tesla's robotaxis struggle in Miami's weather, it won't be a theoretical problem. It'll be a safety problem on public roads with real passengers inside. This is the part where I expected to find a clear answer and didn't.

Waymo has published the most comprehensive safety data of any autonomous vehicle company. They've logged over 100 million fully autonomous miles on public roads and reported over 80 percent fewer injury-causing crashes per mile than human drivers. That's a significant number with a large dataset behind it, and it's the strongest evidence that exists right now that autonomous vehicles can be safer than humans in certain operating conditions.

Tesla claims that FSD is safer than human driving, but the data they publish is structured differently and isn't directly comparable. Tesla's safety reports combine miles driven under Autopilot, which is a highway lane-keeping system, and FSD, which handles city streets. Highway driving is where humans are already safest, so mixing highway miles into the dataset makes the overall numbers look better. Tesla also only counts events where an airbag or seat-belt pretensioner deploys, which leaves out fender-benders, curb strikes, and other incidents that Waymo's data includes. Without equivalent datasets measured in the same way under the same conditions, there's no true apples-to-apples safety comparison between the two systems right now.

The honest answer is that Waymo has demonstrated, with published data, that its system is significantly safer than human drivers in the cities where it operates. Tesla has not yet demonstrated the same thing with comparable data, though the technology is newer and the fleet is much smaller. What is clear is that human drivers are not a great benchmark to begin with: roughly 40,000 people die in car crashes in the United States every year, the overwhelming majority from human error. The bar for autonomous vehicles to be safer than humans is high in perception but surprisingly achievable in practice, because humans are terrible drivers who get distracted, drive drunk, fall asleep, text, and make split-second mistakes that a computer would not. Tesla's long-term vision isn't really about operating a fleet of robotaxis the way Waymo does. It's about turning every Tesla on the road into a potential robotaxi.

The pitch, which Musk has been making since at least 2016, is that your personal Tesla will eventually be able to drive itself without any human supervision, and when you're not using it, you can add it to a shared ride-hailing network where it earns money for you while you sleep, work, or do anything else. Instead of your car sitting in a parking lot depreciating for 95 percent of its life, which is what most cars do, it would be generating income as an autonomous taxi. Musk has said the operating cost would be under twenty cents per mile, compared to two to three dollars per mile for Uber or Lyft. If that math works, it transforms car ownership from a depreciating expense into an appreciating asset.

Tesla is also building the Cybercab, a purpose-built robotaxi with no steering wheel and no pedals, targeting a price under $30,000. Initial production started in early 2026. If the Cybercab works at scale, Tesla's argument is that the cost per mile will be so low that it will be cheaper than taking a bus, which would fundamentally reshape urban transportation.

The caveat is that Musk has been promising full autonomy "next year" for roughly a decade. In 2016 he said all Tesla vehicles would be capable of full self-driving. In 2019 he said there would be a million robotaxis on the road by 2020. In 2022 he said it was coming "this year." The timeline has slipped repeatedly, and each prediction has been wrong by years. The technology is real and advancing, but the gap between Musk's public promises and the actual state of the software has been one of the most consistent patterns in the company's history. Whether the vision arrives in 2027 or 2035 or never determines whether Tesla is worth its current valuation or whether it's the most expensive broken promise in automotive history. So if this comes up in conversation, here's how to think about it. There are cars driving around five American cities right now with nobody inside them. Tesla's robotaxi runs on the same camera-only FSD software that consumer Teslas use, just without a human to catch the mistakes. Waymo uses 29 cameras, five LiDAR units, and six radar sensors per vehicle, and has logged over 100 million autonomous miles with 80 percent fewer injuries than human drivers. Tesla has 42 vehicles in Texas. Waymo has 577 and delivers 250,000 trips a week. Federal regulators have flagged Tesla's camera system for failing in rain and glare. The long-term Tesla vision is that your personal car becomes a robotaxi when you're not using it, earning money at twenty cents per mile. Musk has been promising full autonomy "next year" since 2016. The technology is advancing but the timeline has been wrong every time. Forty thousand Americans die in car crashes every year from human error, so the bar for autonomous vehicles to be an improvement is lower than most people think. Whether we get there through cameras or LiDAR or both is a billion-dollar question that Miami's rainy season might help answer.

Stay informed, stay curious, and we'll see you tomorrow.