Robotaxis: what actually ships (and what still needs a person)
Most "robotaxi" headlines are three different products wearing the same word. One is a car with no driver that you can actually hail. One is a supervised Tesla with a marketing name. One is Uber or Lyft routing you into someone else's autonomous fleet. I care about which product showed up on the street, not which CEO said "this year."
Waymo is the boring answer
If you want unsupervised robotaxi service that already works, Waymo is the company doing it. They spent years on permits, burned through the Cruise-scare regulatory hangover, and now run six-figure weekly ride volumes across multiple cities. The Swiss Re work on their safety record is the kind of thing insurers actually read — I covered that when they claimed better-than-human crash rates.
They also expand by plugging into distribution they don't own. The Uber partnership into Austin and Atlanta is the pattern: Waymo drives, Uber fills the app. That keeps Waymo from having to win consumer brand preference city by city.
Tesla's "Robotaxi" is still a person in the front seat
Tesla's Bay Area launch is useful and also slightly dishonest as branding. The stack is FSD. The product, legally, is ride-hailing with a safety driver because California hasn't cleared unsupervised paid AV service for them. CPUC said they hadn't even applied for the right permits around launch week. So the interesting tech story and the boring compliance story don't match.
I drive a Model 3. I believe the hardware and software trajectory is real. I don't believe "Robotaxi" means empty car until a regulator says so. Watch the permit, not the keynote.
Uber and Lyft buy autonomy instead of building it
Uber's strategy is a portfolio of other people's cars. Lucid and Nuro money, May Mobility, Waymo in selected cities — same idea from different angles. Spend margin now so they still have a marketplace when drivers get expensive or scarce.
Lyft does a thinner version of the same thing: Waymo in Nashville, May Mobility in Atlanta, Tensor Auto later. They're late and smaller. The free-cash-flow story in earnings is how they buy time; autonomy is how they avoid becoming pure labor arbitrage forever. That thread lives more in how gig platforms actually make money.
The unglamorous failure modes
Cruise's San Francisco mess is still the reason regulators flinch. Motional pausing deployments is the other mode: partnerships go quiet when the unit economics or safety case doesn't land.
Then there's the stuff that never makes a keynote. Waymo paying DoorDash drivers ~$11 to close doors is the clearest picture of 2026 autonomy I've written: the car can drive; the operation still needs a human for dumb hardware edge cases. That's not "failure of AI." It's what happens when the expensive robot meets a hinge.
Delivery is a parallel track
DoorDash is running the same autonomy thesis on sidewalks and curb: Serve Robotics, their own Dot robot, Waymo in Phoenix for autonomous delivery. Different vehicle, same labor math — replace the expensive minute, keep a person for whatever the robot can't finish.
Amazon's Zoox in Las Vegas and the Hayward factory story are the vertical-integration bet: own the weird-looking vehicle end to end. Different risk profile than Waymo-on-Jaguar or Tesla-on-Model-Y.
What I actually track
- Unsupervised paid rides in a real city (Waymo yes; Tesla not yet in CA).
- Who owns the customer (app) vs who owns the vehicle (stack).
- Whether human "helpers" show up as a product feature (doors, tows, remote assist).
Everything else is earnings color or press-cycle noise. When a new partnership drops, I ask which of those three moved — and I write the dated post. This page is the map those posts hang on.