Full Self-Driving and the Death of Human Drivers

Drivers

If The Convergence is the thesis of this book, then Full Self-Driving is the

proof of concept. It's the first massive demonstration that AI can replace human labor at scale. Not in a lab. Not in a controlled experiment. On actual roads, with actual people, making actual decisions at 70 miles per hour. And what we're learning from FSD is about to happen everywhere else.

A Disclosure Before We Begin

Full disclosure: I'm heavily invested in Tesla. That creates obvious bias. I own the stock, I drive the car, I use FSD every day. When I tell you Tesla is going to win the autonomous driving race, you should read that with appropriate skepticism. I've tried to be rigorous in my analysis. I've tried to steelman the counterarguments. But I'm human, and humans are notoriously bad at being objective about their investments. So take what follows with that context - and then decide for yourself whether the logic holds. Now let me make my case.

The Experiment That's Already Running

Let’s set a scene. Right now, as I write this in early 2026, there are millions of Tesla vehicles on roads around the world collecting data every single day. Every time a car navigates an intersection, every time it handles a construction zone, every time it deals with a pedestrian doing something unexpected - that data flows back to Tesla's servers. The neural network learns. It gets better. This is literally happening right now. You might own a Tesla yourself (honestly, you probably do if you bought this book). But the scale of this experiment is something that I think most people haven't fully internalized. Tesla can make two million cars per year, and increasing. Not all of them are running FSD at any given moment, but enough are that the company has accumulated billions of miles of real-world driving data. Edge cases that might take a smaller operation decades to encounter? Tesla's fleet sees them daily. Hourly. Minute by minute. Compare this to Waymo, which operates approximately 3,000 vehicles. And I want to be genuinely fair here because Waymo has achieved something remarkable. Within their operating domains, they've demonstrated true Level 4 autonomy - no safety driver, no supervision, just the car handling everything. Their safety record in those areas is impressive by any measure. People take Waymo rides every day and arrive safely. That's not nothing. That's a genuine technical achievement. Even with their mistakes. It’s a clear safety improvement over human drivers, and they should be celebrated. Loudly. But the approach is fundamentally different from Tesla's - and I'd argue fundamentally limited for scale. You’ll quickly see why this means that Tesla’s ability to be everywhere quickly will be such a shock to the current system. Waymo uses high-definition maps of specific areas. Their vehicles know every curb, every lane marking, every traffic light in their operating zones down to the centimeter. Within those zones, the system is brilliant. The

redundant sensor suite - lidar, radar, cameras all working together - creates a level of situational awareness that's genuinely impressive. The limitation shows up when you want to expand. Each new city requires comprehensive mapping before deployment. It's a rules-based system that relies on having complete information about the environment in advance. Tesla's approach is different. They're training a neural network to understand driving the way humans understand driving - by looking at the world and making decisions based on what it sees. No pre-mapping required. The same system that works in Austin works in San Francisco works in Miami works in rural Montana. Because it's not relying on knowing every road ahead of time. It's actually relying on understanding what roads are and how to navigate them. This is the end-to-end neural network approach versus the rules-based approach. That’s a fancy way of saying it’s 100% AI. And from my perspective, there's really no contest in terms of which one scales.

The Math That Most People Miss

I've had this conversation probably a thousand times with people who are

skeptical of Tesla's self-driving ambitions. They point to Waymo's current Robotaxi service as evidence that someone else is ahead. "Waymo has Robotaxis operating right now. Tesla is still mostly in supervised mode." Fair point. As of this writing, Waymo does have a functioning Robotaxi service in a handful of cities. Tesla's unsupervised Robotaxi network is just beginning to roll out. If you're scoring based on who has Robotaxis operating today, Waymo wins. But that's not the right question. The right question is: who can scale? Let me show you the math that makes this undeniable. Waymo can produce somewhere in the neighborhood of 10,000 vehicles per year when they're running at capacity after their partnership with Zeekr (a Chinese EV makers). Their actual fleet is much smaller. Tesla can produce

two million per year, with plans to grow. That's a 200x difference in manufacturing capacity, at minimum. Now think about what that means for a Robotaxi network. Network effects matter in this business. You want enough vehicles that wait times are short, that coverage is broad, that people can rely on the service being available when they need it. Building that kind of network requires massive fleet deployment. With a 200x manufacturing advantage, Tesla can put more Robotaxis on the road in a few months than Waymo can in years. Tesla can put Waymo’s entire fleet on the road in about a day. They can expand to new cities faster. They can respond to demand surges more effectively. They can replace vehicles that need service without significant downtime. And this is before we even get to cost. Waymo's vehicles are heavily customized with expensive sensor packages. Tesla's approach uses vision- based AI running on hardware that's integrated into every vehicle they already make. The cost structure is completely different. But the most important insight - the one that most analysts completely miss - goes deeper.

The Car Sales Hedge

Even if the Robotaxi network scales slowly - even if regulatory approval takes

years, even if there are setbacks and delays - Tesla still wins. Why? Because they can sell cars that drive themselves without anyone paying attention. Consider the implications of that. You're in the market for a car. You can buy a regular vehicle from some legacy automaker, or you can buy a Tesla where the car drives itself on the highway, navigates city traffic, parks itself, and basically handles 99.99999% of your driving without you touching the wheel.

That's a ludicrous competitive advantage for selling vehicles, regardless of

whether there's ever a Robotaxi network. And a gigantic precedent that says that the human driver is no longer needed. Anywhere. The FSD subscription alone adds meaningful revenue - currently $99 per month or a one-time purchase of $8,000, with about 1 million subscribers world-wide out of Tesla’s 9 million car fleet (per Tesla’s Q4 2025 earnings). But more importantly, it's a differentiator that no other automaker can match. When was the last time GM or Ford shipped an over-the-air update that made your car significantly more capable than when you bought it? Never. Tesla vehicles appreciate in capability over time. The neural network gets better, and every car in the fleet gets the update. The car you buy today will be a better driver in a year than it is now. Try finding that value proposition anywhere else. So the Robotaxi opportunity is enormous - I'll get to the numbers in a minute - but even if it takes longer than expected to materialize, Tesla is still selling millions of vehicles with technology that competitors can't match. They accrue miles and data regardless. They improve regardless. They win regardless. This is why I'm not particularly worried about the Robotaxi timeline debates. Whether unsupervised FSD scales up in 2026 or 2028, the underlying trajectory doesn't change. Tesla is building an insurmountable lead in real- world driving data while also selling more cars than their competitors can. And once the regulations catch up, all Tesla has to do is… the same thing. They’ll just tell the drivers “hey so, since your car can already drive itself… would you like it to just make you money by going out and Robotaxi-ing around?” That’s in their literal Master Plan Part 2. The technology is independent from the regulatory environment. If the technology works, then the value proposition of a driver is already obsolete.

The Robotaxi Economics

Now let's talk about what happens when the Robotaxi network does scale, because the economics are frankly staggering. I've seen estimates ranging all over the place, but the consistent theme across my analysis is this: the cost to operate a self-driving vehicle is going to plummet to somewhere around 30 cents per mile - probably lower. A typical Uber ride costs passengers $2-3 per mile in most markets, with the driver taking a cut and Uber taking a cut. The fully-loaded cost of operating that ride - including the driver's time, vehicle depreciation, fuel, insurance, everything - is roughly $2.80 per mile on average in the US. Much higher in dense urban areas like New York or San Francisco. Now you have a Tesla Robotaxi operating at 30 cents per mile. Even if Tesla charges $1.50 per mile to passengers - significantly cheaper than Uber - they're making over a dollar of margin on every mile driven. At scale, with millions of vehicles operating around the clock, the numbers become astronomical. And unlike human drivers, these vehicles don't need breaks. They don't have shifts. They can operate 20+ hours per day with only charging downtime. A single Robotaxi could replace three to five Uber drivers in terms of passenger-miles served. The cost per mile advantage isn't incremental. It's transformational. This is why I've said in the past that Uber is completely and utterly cooked by this dynamic. Not damaged. Not hurt. Cooked.

The Cybercab Factor

Now, I should mention the Cybercab specifically, because it's going to

accelerate these economics even further. Tesla's current Robotaxi deployment uses existing vehicles - Model Ys, cars that were designed for human drivers. They work fine for this purpose, but they're not optimized for it. They have steering wheels and pedals that

nobody in a Robotaxi needs. They're designed for a use case that doesn't apply. The Cybercab is Tesla's purpose-built Robotaxi vehicle. Two seats facing forward. No steering wheel. No pedals. Compact form factor. Top speed of 100 miles per hour. Giant screen. Designed from the ground up for autonomous passenger transport. 90% of rides in the US are 2 people or less. You see what I’m saying? Why does this matter? Because purpose-built is always cheaper and better than retrofitted. A Cybercab will cost significantly less to manufacture than a Model 3 because there's less stuff in it. No steering column, no brake pedal, no mirror adjustments, no controls that a human driver would need. You can fit more of them in a manufacturing line. The bill of materials is simpler. And because it's designed for Robotaxi use, the interior can be optimized for passengers. More comfortable seating. Better placement of screens and entertainment. Easier entry and exit. All the things you want when you're designing a car that will transport thousands of different passengers rather than one owner. When Cybercab production ramps - and Tesla is targeting somewhere around 800,000 units per year capacity by 2027 based on my calculations - the cost per mile will drop even further. I've seen projections as low as 25 cents per mile or even sub-20 cents at scale. At those numbers, Robotaxi rides become cheaper than owning a car for a lot of people. Think about what that does to car ownership patterns. If hailing a Robotaxi is cheaper than the monthly cost of car payments, insurance, maintenance, and fuel combined - why would you own a car? Especially if you live in an urban area where parking is a nightmare anyway? This is why I think the long-term impact of Robotaxis goes way beyond ride- hailing. It changes the entire structure of personal transportation. The second-order effects are massive.

What Happens to Uber

I should probably expand on the “Uber is cooked” bit, because it sounds

extreme. And look, Uber is a multi-hundred-billion-dollar company. They're not going to disappear overnight. Their CEO is my Persian Brother from another mother. But their current business model is fundamentally uncompetitive with Robotaxi economics. Uber's entire value proposition is connecting riders with drivers. They take a percentage of each ride for providing that connection. But what happens when you don't need drivers? You might think Uber could just add self-driving vehicles to their platform. And they’re already trying it with Waymo. In Austin, the only way to get a Waymo self-driving car is through Uber. But this exposes a core vulnerability: Uber doesn't manufacture vehicles. They don't develop autonomous driving software. They're a matchmaking service. If Tesla operates its own Robotaxi network - which is exactly what they're doing - why would Tesla let Uber take a cut? Tesla can go direct to consumers. Tesla can capture the entire margin. Uber becomes an unnecessary middleman. The only scenario where Uber survives in recognizable form is if they become a platform that aggregates across multiple Robotaxi providers. But even then, they're fighting for a slice of a pie that's dramatically smaller than the current human-driver model generates. Tesla’s fleet will be 100x larger than everyone else combined for the foreseeable future. How do you compete with that? The platform fee on a Robotaxi ride isn't going to be 25% like it is with human drivers - there's no driver to subsidize or recruit. So Uber loses on cost AND scale. I've heard the argument that Uber has brand recognition and customer relationships. Sure. But customers have relationships with lots of companies that got disrupted. People had relationships with Blockbuster too. Brand loyalty doesn't mean much when someone else offers a better product at half the price.

The same logic applies to Lyft, though honestly Lyft was already struggling to

achieve real profitability even before Robotaxis. The transition is going to be painful for both, but the direction is clear.

The Regulatory Reality

Now I want to address something that comes up constantly in discussions

about self-driving vehicles: regulation. And the argument goes like this: "Even if the technology works, regulators will never approve truly unsupervised Robotaxis. There will be endless bureaucratic delays. Liability issues will be impossible to resolve. Tesla's timeline is wildly optimistic." I'm not going to pretend that regulatory approval is simple. It's not. Different states have different rules. Some jurisdictions will be faster than others. There will absolutely be friction. But I think people overweight the regulatory obstacle and underweight the technological one. The technology is the hard part. The technology is what people said would never work. And the technology is increasingly working. Once you have self-driving vehicles that are demonstrably safer than human drivers - and the data is showing exactly that - the regulatory argument shifts from "should we allow this?" to "can we afford not to?" 40,000 people die in car accidents in the United States every year. The vast majority of those accidents are caused by human error. If AI systems can reduce that number significantly - and FSD safety statistics suggest they can - then blocking autonomous vehicles becomes a choice to let people die. That's a hard position for regulators to defend indefinitely. Eventually, they will have to cave. Not to mention the massive economic benefits of having cheap, reliable, safe point to point transport at your fingertips. I think what we'll see is a patchwork rollout. Some states will approve quickly. Others will lag. Eventually the successes will create pressure on the

holdouts. The same pattern we saw with EV adoption generally. Will definitely be slower, but it’s inevitable. Austin is already live. California is happening. Arizona has been approving expansions. The momentum is building. Anyone betting that regulation permanently blocks this is betting against a force that saves lives and reduces costs. I wouldn't take that bet.

The Safety Argument

I want to spend a moment on the safety statistics because they're often

underappreciated in this debate. Tesla publishes quarterly safety reports comparing FSD and Autopilot to both Tesla vehicles with no automation engaged and to national averages. The numbers are stark. Tesla vehicles using FSD or Autopilot have significantly fewer accidents per million miles than human drivers. This is not just a Tesla phenomenon; it’s a self-driving one. Waymo has similar statistics for their operations. Now, skeptics will point out various methodological issues with these comparisons. FSD is mostly used on highways and in good conditions. Human accident rates include all driving, including challenging situations. The comparisons aren't perfectly apples-to-apples. Fair enough. But even accounting for methodology, the trend line is clear: as the AI gets better, the safety gap widens. Every version of FSD is safer than the last. The trajectory points toward a future where letting humans drive is the risky choice, not the safe one. And this creates real moral pressure on regulators. How do you tell a family that lost someone in a car accident that you're blocking a technology that could have saved their loved one? How do you justify protecting jobs when the cost is measured in human lives? These AI systems will save lives. Regardless of what you think of Tesla as a company, regardless of what you think of Elon Musk as a person, it is an

objective fact that this technology will reduce deaths from auto accidents once it's widely deployed. The only question is how quickly we get there. Some people argue that we should wait until the technology is absolutely perfect before deploying it. But perfect is the enemy of good. Human drivers kill over 40,000 people per year in the United States. A self-driving system that's merely twice as safe as humans would cut that number in half. Should we wait years for perfection while people die in preventable accidents? I don't think so. And increasingly, neither do regulators.

The Technology Behind the Wheel

I want to get a bit more specific about how FSD actually works, because I

think understanding the technology helps explain why I'm so confident in its trajectory. Traditional approaches to autonomous driving involved writing rules. Lots of rules. If this happens, do that. If you see a stop sign, stop. If a pedestrian is in the crosswalk, wait. If the light is green and the intersection is clear, proceed. The problem with rules-based systems is that driving involves infinite situations. You can't write a rule for everything. What happens when there's a construction zone with a worker waving you through a red light? What happens when the lane markings are covered by snow? What happens when someone does something completely unexpected? Rules-based systems handle edge cases poorly because every edge case requires a new rule. And the rules have to interact correctly with all the other rules. The complexity explodes. Tesla's end-to-end neural network approach is fundamentally different. Instead of writing rules, they train a neural network on actual driving behavior. The network sees what the world looks like through cameras. It sees what human drivers do in response. It learns the mapping between visual input and appropriate action.

This is more similar to how humans actually learn to drive. Nobody hands

you a rule book with ten million entries. You watch, you practice, you develop intuition. The neural network does something analogous, except it can learn from millions of drivers simultaneously. And the thing about this approach is that it scales differently. More data makes the network better. More edge cases just become more training examples. As the system handles more situations, it simply gets more capable without adding complexity. And because every Tesla vehicle is potentially collecting data, the training set grows constantly. The network sees rare situations across the entire fleet and learns from them. An edge case that any individual driver might encounter once in a lifetime, the fleet encounters hundreds of times per day. And the comparison to Waymo goes beyond manufacturing scale. It comes down to data scale and approach. We're talking about building a system that fundamentally improves with size rather than one that requires more complexity to handle more situations.

The V14 Inflection Point

So now I want to get specific about the current state of FSD, because I think

we're at the inflection point where The Convergence thesis starts to play out. Version 14 of Tesla's FSD stack, which has been rolling out to vehicles, represents a significant leap in capability. The smoothness of interventions, the handling of complex scenarios, the confidence of the system - it's dramatically better than what we saw even a year ago. I've been doing FSD drives regularly and tracking the improvement on my YouChannel (Farzad, like and subscribe). The difference between V12 and V14 is like night and day. V12 was impressive but clearly needed babysitting. V14 feels like a competent driver that rarely needs a correction - if ever. And those corrections are not safety related - they come down to preference. “Oh it should’ve moved over just a tad bit faster”. “Eh it’s waiting here too long”. Stuff like that. The trajectory toward truly unsupervised operation is clear.

And this is with hardware that's been in vehicles for years. When AI5 -

Tesla's next-generation compute hardware - begins deploying, the system will

have significantly more processing power to work with. More compute means more capability, means handling more complex scenarios, means fewer edge cases where human intervention is needed. The progress isn't linear. It's been accelerating. Each version improves faster than the last because the neural network is getting smarter at learning from data. The flywheel from Chapter 1 applies directly here: more data makes the AI better, better AI collects more useful data, the cycle accelerates. I think that within the next three to nine months from the date of publishing this book (February 14th, 2026), we're going to see a storm of self-driving Teslas operating in metropolitan areas that approve the technology. At that point, the scale advantage becomes impossible to ignore.

The Human Driver Endgame

I want to zoom out and connect this back to the theme of this book. Full Self-Driving represents something much bigger than Tesla, Robotaxis, or rideshare market share. What matters most is what happens when AI can replace human labor in a major category. Driving is one of the most common occupations in many countries. In the United States alone, there are roughly 3.5 million professional truck drivers, millions of taxi and rideshare drivers, delivery drivers, bus drivers. Driving- related occupations collectively employ a lot of Americans - and we’re not even talking about those of us that drive because we want to. Or because we have to. FSD represents the first wave of AI in the physical world that's going to displace jobs immediately and at scale. Not gradually over decades. Over years. This is going to bring about amazing safety improvements. The reduction in accidents alone will save tens of thousands of lives per year once fully

deployed. The cost per mile for transportation will drop dramatically, making mobility more accessible to everyone. These are genuine benefits that will improve people's lives. But it's also going to eliminate jobs. Real jobs held by real people who currently make their living driving. And unlike previous technological disruptions where new jobs emerged to replace old ones, I'm not sure what driving-equivalent employment emerges for displaced drivers when AI is deployed everywhere digitally, and robots start becoming real in the physical world. This is the pattern we're going to see repeated across industry after industry as AI capability grows. Tremendous benefits for consumers and capital owners. Tremendous displacement for workers in affected categories. Don’t think about the technology itself being either good or bad - it's a forcing function that creates both outcomes simultaneously. I'll discuss what this means for different segments of the population in Part II of this book. But I wanted to flag it here because FSD is the proof of concept. Whatever you think is going to happen with AI and labor broadly, FSD is the canary in the coal mine. It's showing us exactly how this plays out.

The Scalability Question

Now, I want to return to the competitive question one more time, and I want

to be intellectually honest about where I could be wrong. The case for Waymo is stronger than Tesla bulls often admit. They have working Robotaxis operating right now - not supervised, not beta, genuinely autonomous. They're owned by Google, which has deep pockets, world-class AI talent, and patience measured in decades. Their safety record within operating domains is excellent. Their technology works. If I'm wrong about scalability mattering more than early technical leads, Waymo could prove to be the better approach. Maybe mapping gets dramatically cheaper and faster. Maybe they find manufacturing partners who can produce at scale. Maybe the geofenced model expands city by city until

it's effectively everywhere. I'm betting they can't scale fast enough, but I've been wrong before. My thesis anyway: I think the winner is determined by who can deploy everywhere, not who achieves perfect operation somewhere. Working Robotaxis in geofenced areas are impressive. But translating that to nationwide coverage requires either mapping everywhere or developing the kind of generalizable AI that Tesla is building. Google's deep pockets can fund either approach - but money can't buy the billions of real-world training miles that Tesla's fleet generates every day. Waymo's path to matching Tesla's scale requires either: building manufacturing capability from essentially zero to millions of units per year, or licensing their technology to automakers who can manufacture at scale. Option one would take years and billions of dollars. The manufacturing learning curve is steep, and even Google's resources don't shortcut the time required to build production expertise. Option two runs into partnership challenges. Any automaker who licenses Waymo's technology becomes dependent on a partner. If you're Toyota or Ford, do you want Google as a gatekeeper for your vehicle's most critical feature? Some might accept that tradeoff. Many won't. The same logic applies to GM's Cruise, to Apple's abandoned car project, to anyone else trying to solve autonomous driving. The data advantage from having millions of vehicles on the road, constantly training, is almost impossible to replicate from behind. Some people have suggested that Chinese companies like Huawei or BYD might catch up through sheer scale of the Chinese market. And I take that seriously - China shouldn't be underestimated, and I'll discuss the US-China competition in depth later. But even Chinese manufacturers face the challenge of building end-to-end neural network capability from scratch. They can see what Tesla has achieved; replicating it is much harder.

The Timeline Debate

I want to address the timeline question directly because I know it's where

skepticism is highest. Predictions about autonomous vehicles have been notoriously wrong. Companies have promised Robotaxis for years and under-delivered. Every timeline Elon Musk has given for FSD has slipped. The skeptical position is that this pattern will continue indefinitely. I understand this skepticism. And I'd acknowledge that specific date predictions are difficult. When I say "within months" or "by 2027," I'm giving my best estimate, not a guarantee. But I'd push back on the idea that past delays mean the technology never arrives. The difference between "taking longer than predicted" and "never happening" is enormous. And the technology is clearly arriving. V14 is demonstrably close to unsupervised capability. Regulatory approvals are accumulating. Robotaxi networks are launching. The question isn't whether self-driving cars work. They work. The question is how quickly deployment scales. And on that question, I think the skeptics are underestimating how fast this can move once the regulatory gates open. Consider this: right now Tesla has millions of vehicles that could potentially operate as Robotaxis with a software update. Not future vehicles. Current vehicles. When approval comes for unsupervised operation in major markets, deployment doesn't require building new factories or manufacturing new cars. It requires flipping a switch. That's unprecedented in technology deployment. Usually scaling requires building things. In this case, the things are already built and waiting.

Where does this leave us?

Full Self-Driving is the proof point for everything I discussed in Chapter 1. It demonstrates that AI can learn complex real-world tasks. That data advantage compounds over time. That vertical integration - designing the car, the

software, the neural network, the compute hardware, the manufacturing - creates advantages that horizontal players can't match. And it shows us what the transition looks like in practice. Revolutionary technology that genuinely improves lives. Massive disruption to existing industries. Jobs eliminated. Wealth created. A forcing function for societal change. This is a microcosm of the abundance or collapse choice. FSD success means cheaper, safer transportation for everyone - a step toward abundance. But it also displaces millions of drivers. How we handle that displacement determines whether this technology becomes a blessing or a source of social chaos. The same pattern will repeat across every industry The Convergence touches. Robotaxis are coming. The death of human drivers as a mass profession is not a question of if but when. And "when" is measured in years, not decades. In the next chapter, I'm going to talk about Optimus - Tesla's humanoid robot program. Because if you think replacing human drivers is transformational, wait until you see what happens when robots can replace human labor in virtually any physical task. The flywheel is just getting started. And so is the disruption.