The Meter,
the App &
the Machine

Here is the loss nobody put on a balance sheet. Driving a cab used to be a job — with everything that word carries. You came to the garage. You picked up your car from the same bay every morning. There was a shift, which meant there was a start and an end. There was a dispatcher who yelled at you, which meant there was someone who knew your name. And when you clocked out, you clocked out with the other people who had just done the same thing, and you went and got a drink and complained about your day — exactly like a stockbroker, a framer, a paralegal, or a cop.

That is not sentimentality; it's structure. The cab garage was a workplace, and workplaces manufacture something beyond their product: they make colleagues.

The culture noticed. Taxi — the sitcom that ran on ABC and NBC from 1978 to 1983 — was not set in a cab. It was set in the Sunshine Cab Company's Manhattan fleet garage, and the show worked because the garage was where the drivers actually lived, needled each other, and got berated by Danny DeVito's dispatcher from inside his cage. It won 18 Emmy Awards, including three straight for Outstanding Comedy Series, and it was built from a 1975 New York magazine piece about the night shift at a real Greenwich Village garage. Whole films took the front seat as their subject: Scorsese's Taxi Driver (1976), Michael Mann's Collateral (2004). The cab was a place where American stories happened.

And then there's the one that proves the other half of this article's argument almost too neatly.

Taxicab Confessions premiered in 1995 and ran, on and off, for over a decade. The entire premise was that under the cover of a city night, in a vehicle with a stranger who would never see you again, people will say anything. That show is a documentary record of a psychological condition that has since been engineered out of existence — and we'll come back to it.

What rideshare gave the driver

To be fair — and this matters — the trade was not one-sided. Rideshare handed drivers something the garage never did: total autonomy. You set your own hours. You work Tuesday and skip Wednesday. Nobody assigns you a weekend, refuses your overtime, or holds a shift over your head. For a parent, a student, a retiree, or someone bridging between careers, that flexibility is not a consolation prize — it's the entire appeal, and it is why millions of people took the work.

What they gave up was everything on the other side of the ledger: no coworkers, no employer, no benefits, no bargaining power, no floor under the fare, and — for the first decade — no legal path to organize at all, because independent contractors sit outside the federal labor law that gives employees the right to unionize.

That last part has started to change, and it's worth being precise about it, because the picture is genuinely mixed.

• 2018 · New York City First major U.S. city to cap new ride-hail licenses and impose a driver minimum-pay floor (~$17.22/hour).
• 2020 · California Proposition 22 passes, keeping app drivers classified as independent contractors while adding limited earnings and healthcare guarantees.
• June 2024 · Massachusetts Uber and Lyft settle with the state attorney general for $175 million, agreeing to a $32.50 hourly minimum, sick time, and a health-insurance stipend.
• November 2024 · Massachusetts Ballot Question 3 passes with 53.9% — the first law in the nation giving rideshare drivers the right to form a union and bargain across an entire industry. Roughly 70,000 drivers are covered. They remain independent contractors, and delivery drivers are excluded.
• 2025–2026 · Everywhere The robotaxi arrives, and the question of what the driver is owed becomes, in a growing number of trips, moot.

So the arc runs: the cab driver had colleagues and a union and a boss. The rideshare driver traded all three for a schedule of their own and is now, city by city, slowly clawing back a fraction of what was traded. And the robotaxi resolves the whole labor question by deleting its subject. Waymo's car doesn't negotiate, doesn't organize, doesn't clock out and doesn't go for a drink. It works until the battery needs charging, and then it works again.

Now the obvious question, and the one that gives this piece its charge: is Waymo about to do to Uber and Lyft exactly what Uber and Lyft did to the taxi? Remove the human, undercut the incumbent, flip the market?

The raw numbers say a revolution is underway. Waymo's paid rides went from about 10,000 a week in 2023 to 100,000 by August 2024, past 200,000 by early 2025, and — by analyst counts in early 2026 — roughly 400,000+ a week across six cities. It logged around 15 million rides in 2025 alone, tripling year over year. In San Francisco it has reportedly climbed to 25–27% market share, surpassing Lyft inside its service area. In February 2026 it raised $16 billion at a $126-billion valuation, with expansion planned to 20-plus cities including Tokyo and London.

But here's the twist

So far, robotaxis are not cannibalizing rideshare the way rideshare cannibalized cabs. The counterintuitive data point of 2025: Uber and Lyft have been growing roughly five times faster in cities with robotaxi fleets than in cities without them. Lyft's CEO calls autonomy a "massive TAM expander" — driverless cars are pulling people out of their own cars and into on-demand rides generally, and Uber is smart enough to be partnering with Waymo rather than fighting it, dispatching Waymos through the Uber app in Austin and Atlanta.

Why the difference from the taxi bloodbath? Because Waymo's constraint isn't demand — it's supply. Each car costs a fortune, the fleets are small, and the service still skips airports and highways. Uber didn't have to build a car; Waymo has to build every single one. For now, the pie is growing faster than the robots can eat it.

The other bet

Which is exactly why Tesla matters to this story, and why it deserves more than a footnote. Tesla is running the opposite wager, and the two approaches disagree about the most fundamental question in the industry: what a self-driving car needs in order to see.

Waymo's answer is everything — LiDAR, radar, cameras, and centimeter-accurate maps of every street it operates on, scanned block by block before a single paying passenger gets in. That's why each car is expensive and each new city takes months. Tesla's answer is cameras only. No LiDAR, no pre-built map. If it works, the economics invert completely: expanding a service area becomes close to a software toggle, the hardware is a Model Y instead of a sensor-crowned research vehicle, and the potential fleet is the millions of FSD-capable Teslas already sitting in driveways waiting for their owners to opt in. Waymo has to manufacture its supply. Tesla, in theory, already sold it.

That is a genuinely enormous "if," and as of mid-2026 it remains unresolved.

Tesla launched its robotaxi service in Austin in June 2025 with roughly ten to twenty Model Ys and a safety monitor riding in the passenger seat. The regulatory door opened on May 28, 2026, when a new Texas driverless-vehicle law took effect and Tesla self-certified at Level 4 the same day. On June 3 it removed the monitors and expanded unsupervised coverage to the entire Austin metro — roughly 245 square miles, airport included. On the map, Tesla had just out-covered Waymo in a major American city.

Then you count the cars. Independent trackers put Tesla's active unsupervised Austin fleet at around 20 vehicles serving all 245 of those square miles. State filings tell the same story from the regulator's side: 42 Tesla robotaxis registered in Texas against Waymo's 577 — the same state, the same paperwork, a 13-to-1 gap. Nationally the asymmetry is starker still. Waymo told NHTSA in December 2025 that it operates 3,067 robotaxis, and it now runs over half a million paid rides a week across ten U.S. cities.

The safety picture is thinner and harder to read. Tesla's Austin operation logged 17 reported incidents with NHTSA between July 2025 and April 2026 — two involving minor injuries, one requiring hospitalization. Electrek, which has tracked the program since launch, calculated a crash rate roughly nine times the human benchmark; with a fleet this small the error bars on any such rate are wide, and Tesla disputes the framing. But note the asymmetry that matters: Waymo's safety case rests on a peer-reviewed study of 56.7 million driverless miles. Tesla's rests on a pilot. Musk has tied real scaling to FSD v15, an architectural rewrite expanding the driving model from roughly one billion parameters to ten, targeted for late 2026 or early 2027.

One more detail belongs here, because it closes a circle this article opened in 1990. Tesla's purpose-built robotaxi, the Cybercab, is a two-seater designed with no steering wheel and no pedals at all. Not a car with the driver removed — a car in which the driver's position was never designed in the first place.

What the Cybercab actually is

Musk unveiled it in October 2024 and said it would sell for under $30,000 — roughly $31,500 in 2026 money. Two seats, on the argument that 90% of miles are driven with one or two people aboard. No steering wheel, no pedals, nothing to adjust. Production began at Giga Texas in April 2026, slowly, on what Musk called a stretched-out S-curve. The first production cars carried paying passengers in Austin on September 4, 2026.

The interesting part isn't the car. It's the paperwork. A vehicle with no steering wheel does not obviously satisfy federal safety standards written on the assumption that somebody is steering. Zoox, Amazon's robotaxi arm, went the formal route — it petitioned for a Part 555 exemption, was granted a temporary waiver from eight standards, and launched commercially in July 2026, bound by the rule capping exempted vehicles at 2,500 units. Tesla didn't petition. Tesla self-certified the Cybercab as compliant with every existing standard, the same process Toyota uses for a Camry, and in doing so stepped around the cap entirely.

The regulator noticed. On the same day the first Cybercabs took passengers, NHTSA opened an investigation into "the process and technical data on which Tesla relied when certifying the Cybercab," noting that existing standards remain in force until that work is finished. Tesla has, in effect, declared its own car legal, and is now being asked to show its working.

The part that sounds like Airbnb

Which brings us to the thing that should interest anybody reading this who has a driveway.

Musk's pitch was never that Tesla builds and runs a fleet the way Waymo does. It's that you do. "I think it would end up being some kind of combination of like Airbnb and Uber," he said — you buy the car, and when you aren't using it, it goes out and works. The arithmetic attached to it is genuinely seductive. A private car sits idle almost all of its life, in use maybe 12 hours a week. On the network, 60. "The interesting thing," Musk noted, "is the car still costs the same." At Tesla's 2019 Autonomy Day he put a number on the upside: around $30,000 a year in gross profit per vehicle, roughly $38,700 in today's money.

On September 3, 2026 — the day before those first Cybercabs took passengers — Tesla published a form asking who wanted in. The options: Cybercab fleet vehicle purchasing, mobility hubs and infrastructure, event collaboration, other.

Read that list again, because the word that matters is the one that isn't on it. The form is addressed to businesses. Not to you.

And there's a harder question underneath the pitch, which Electrek's Fred Lambert has put about as bluntly as it can be put: if a Cybercab on that network genuinely printed $30,000 a year, why would Tesla hand that money to a stranger instead of keeping every car for itself?

That isn't a rhetorical flourish. It's the test. A company that has built a machine which prints money does not sell you the machine. It buys more of them. Waymo — which has the working autonomy and the operating proof — sells nobody a car and operates every one of the 3,067 it told NHTSA it owns. The moment a robotaxi business becomes willing to sell you the vehicle is the moment worth asking what the seller knows about the margin that the buyer doesn't.

The precedent isn't encouraging, because Tesla has made this exact offer before. Owners who bought Full Self-Driving — at up to $15,000, about $16,900 today — were told they would be able to add their cars to something called the Tesla Network and earn taxi income from them. That promise dates to 2016, and was restated in 2019. It has never shipped. Tesla runs the service itself, with its own cars, and the people who paid for FSD on the strength of that revenue are, a decade on, still unable to collect it. The September 2026 form did not change that: individual FSD owners remain left out.

Then there's the asset itself. MisterGreen, a Dutch leasing company, bought more than 4,000 Teslas on precisely this reasoning and lost $40 million when the residuals went against it — used Teslas depreciating at roughly three times the rate of the broader used-car market. A fleet owner isn't only betting on ride demand. They're betting on resale values, on software arriving when it was promised, and on a regulator that is at this moment asking the manufacturer to justify the car's legality.

Strip the technology out and the offer is the oldest one in this story. In 1937 a city sold a man a piece of tin and called it an annuity. In 2012 a platform sold him be your own boss and took a cut that grew every year after. In 2026 the pitch is unchanged — buy the asset, let it earn while you sleep — and the genuinely new part is only that nobody has to sit in the car. Whether the person holding the keys keeps the money has never once been settled by the machine. It has been settled by whoever wrote the terms.

Three questions, three honest answers, no crystal ball required — just the direction the data is already pointing.

The tidy narrative — taxi killed by rideshare, rideshare killed by robot — is too neat. What's actually happening is stranger and more interesting: each layer absorbs the last rather than simply burying it. The medallion became an app; the app is becoming a dispatcher for machines; the machine still runs on the network the app built. The driver is disappearing from the front seat — but the ride, in every era, only gets more essential.

Which returns us, finally, to 2084.

Verhoeven's Johnny Cab was a warning dressed as a gag: the machine will be polite, efficient, and catastrophically literal, and one day you will need it to understand something it cannot. What he could not have predicted is that we'd run the numbers and find the machine crashing 80% less than we do, refusing nobody, robbing nobody, and never once taking the long way. The joke aged into an argument.

But hold onto the last frame anyway. Quaid got out of that cab because he could tear the driver out and put his own hand on the control. There is no joystick under the dash of a Waymo — and the Cybercab is being built without a steering wheel at all. What's left instead is a support line, a fleet-wide software patch, and a company deciding what your car does next. We are not choosing between a human driver and no driver. We are choosing between a driver we can argue with and one we can only file a complaint about.

The meter's still running. It just doesn't need a hand to start it anymore — and it no longer has one to take.