Autonomy stopped promising and started shipping

The autonomous vehicle world spent a decade selling demos. This past week, the story shifted. Tesla is running a Cybercab with no steering wheel. Waymo pulled its cars out of Uber in Phoenix. A robot-hand startup shipped its first batch. And a defense company won a real production contract, not a pilot. Let me catch you up on what actually moved and what it means for how you place your own bets.

The cars are losing their controls

Tesla started testing a production Cybercab in Austin with no steering wheel and no pedals, per TechCrunch's Sean O'Kane. A safety monitor still rides in the passenger seat, but the driver's controls are gone. That is a hardware commitment, not a slide.

What made it possible was a rule change. The National Highway Traffic Safety Administration proposed not requiring brake pedals in cars "designed to be driven exclusively by automated driving systems." It is still in public comment, but expected to pass this year. Regulation is starting to catch up to the product, which removes an excuse everyone leaned on.

The tradeoff is visibility. Gold two-seat Cybercabs stand out. When one does something wrong, everyone sees it, unlike the lightly modified Model Y cars Tesla has been hiding behind.

Waymo doesn't need the middleman anymore

Waymo pulled its robotaxis off Uber's app in Phoenix, ending a three-year partnership. The cars went back into Waymo's own fleet. Uber called it "an intentionally limited deployment" of just over a dozen vehicles. Waymo called it a jumping-off point.

The numbers explain the breakup. Waymo now runs around 4,000 vehicles, operates in 11 U.S. metros, and does more than 500,000 trips a week. When you have that reach, the app that used to bring you riders becomes a competitor. The two are set to go head to head in London this year.

Waymo is also stacking supply. MoffettNathanson dug through shipping records and found Waymo on pace to import 3,156 Zeekr-made vans this year, about 300 a month. That is a company building for volume, not for a demo.

The sensor fight is now a business rule

Lyft CEO David Risher drew a hard line: cars that use only one type of sensor can't run on the Lyft network. Lyft confirmed that rules out Tesla's camera-only Cybercab and its unsupervised robotaxis. Human drivers using Tesla cars are fine, so this is aimed squarely at the autonomy product.

So you have two camps. Tesla wants cameras only, because it builds the car and the software and controls its own costs. Waymo runs 13 cameras, four lidar, six radar, and audio receivers on its newest van. One bet is cheaper to scale. The other is easier to defend when a regulator or a fatal crash lands on your desk.

And crashes are landing. NHTSA and the NTSB both opened investigations into a fatal Texas crash tied to a Tesla, while Tesla settled a separate suit over a 2023 FSD crash. Scrutiny scales with deployment.

Money and metal are moving off the roads too

This is not just robotaxis. Seattle's Overland AI won a $19.7 million Marine Corps contract to build more than a dozen self-driving ground vehicles, with deliveries starting in early 2027. CEO Byron Boots says they are the first ground autonomy company to be the prime contractor on a military production deal, not a supplier buried under someone else. That is the line between a demo and a program of record.

Freight is heating up the same way. Humble Robotics came out of stealth with $24 million to build a cabless electric hauler, founded by Eyal Cohen, an Otto and Pronto veteran. The people who lived through the 2016 hype are building the next round.

Investors are betting on staying power. Wayve launched an $85 million employee tender offer at an $8.5 billion valuation, letting staff cash out vested equity so they don't leave. When a startup pays to keep its people through a long build, it is planning for years, not a quick exit.

The deep cut

The piece that changes your planning is the robot hand. Proception settled its trade-secret fight with Tesla and raised $11 million to build a 22-degree-of-freedom hand, shipping its first batch now. Founder Jay Li wants to be the supplier so other companies don't have to solve dexterous manipulation themselves. A Northwestern robotics director told the Wall Street Journal that useful human-level hands are likely a decade out.

Here is the pattern across all of it. The hard part is being sold as a component. Waymo buys vans and adds software. Overland ships vehicles plus the code that runs them. Proception sells the hand. If autonomy is your roadmap, the question is no longer whether to build the whole stack. It is which layer you buy and which one you own. Betting on owning everything, like Tesla, is a real strategy, but it means you eat every crash and every regulator too.

Three questions for your team

1. If a supplier will sell us the hardest layer as a component, which parts of our stack do we actually need to own, and which are we building out of habit?
2. Our safety story scales with our deployment. If a partner like Lyft or a regulator set a hard technical bar tomorrow, does our product clear it or get cut?
3. Are we structured to survive a multi-year build, on funding and on retaining the people who know the domain, or are we priced for a demo?