The most important technology in the robotaxi race isn’t the car. It’s the paperwork. And the two companies furthest ahead in this contest seem to agree with me, because they’ve been spending like it.
Reports this week say Waymo has doubled its lobbying spending as it squares off with Uber over the future of self-driving taxis. In New York alone, the two companies have poured a combined total of more than $15 million into lobbying state politicians this year. That’s not a rounding error in a marketing budget. That’s a strategy.
If you follow AI mostly through demos and benchmark charts, this might feel like a boring detour. It isn’t. It’s one of the clearest signals we have about how AI agents actually reach the public.
What a robotaxi really is
On this site I usually explain AI agents as software that takes actions on your behalf. It books the flight instead of just telling you which flight to book. It files the ticket instead of drafting an email about the ticket.
A robotaxi is that same idea wearing a much heavier coat. It’s an AI agent with a steering column. It perceives, decides, and acts continuously, in public, at speed, surrounded by people who never agreed to be part of anyone’s product test.
Which means a robotaxi has a requirement that a chatbot does not: permission. A text-generating agent needs a user. A driving agent needs a user, a city, a state regulator, an insurance framework, and a body of law that agrees the thing on the road is allowed to be there. No amount of clever engineering creates that permission. Only politics does.
Why lobbying is a product decision
Here’s the part I find genuinely interesting. When a company doubles its lobbying budget, it’s telling you where it thinks the hard problem lives.
Consider two possible bottlenecks for a robotaxi company. The first is technical: the vehicle can’t handle a particular intersection, a particular weather pattern, a particular unpredictable pedestrian. The second is legal: the vehicle handles all of it beautifully and still isn’t permitted to operate in the city you want.
You solve the first with engineers. You solve the second with lobbyists. And the spending patterns suggest both companies now see the second one as the tighter constraint, at least in specific markets. New York is a useful example because it’s dense, heavily regulated, and has an existing taxi and rideshare system with its own politics and its own constituencies. Getting cleared to operate there is not a software update.
There’s also a competitive wrinkle. Waymo builds and operates its own self-driving vehicles. Uber runs a platform that connects riders with rides. Those are different businesses with different ideal rulebooks. Critics have accused Uber of seeking regulatory arrangements that suit its model. Waymo, unsurprisingly, has its own preferences. When two companies with different shapes both want to write the rules, you get a bidding war on influence rather than a shared industry standard.
The pattern worth remembering
I want to zoom out, because I think this generalizes well beyond taxis.
As AI agents move from giving advice to taking action, they cross from the software world into the regulated world. Suggesting a stock pick is a content problem. Executing a trade is a securities problem. Summarizing a medical paper is a search problem. Recommending a treatment is a licensing problem. Describing a route is a maps problem. Driving it is a transportation problem.
Every one of those crossings comes with an existing rulebook, existing incumbents, and existing lawmakers. So the companies building action-taking agents will increasingly compete on two fronts at once: technical capability and regulatory positioning. The robotaxi fight is simply the version of this that’s furthest along, because cars are physical and visible and the laws around them are old and specific.
For those of us watching from outside, that suggests a useful habit. When you want to know how close an AI agent is to reaching you, don’t only ask whether it works. Ask whether it’s allowed. Those are separate timelines, and the second one is often slower.
What I’d keep an eye on
- Where the money goes, not just how much. State legislatures and city councils are where operating permissions get decided. Spending concentrated there tells you which markets a company considers winnable.
- Whose rulebook wins. Rules shaped around a vehicle operator look different from rules shaped around a platform. That difference determines who can enter the market later.
- Whether riders and residents get a voice. Two companies with large budgets are not the only stakeholders on a public street, even if they’re currently the loudest.
The engineering story of self-driving cars has been told for years. The governance story is younger, messier, and arguably more decisive. A doubled lobbying budget is a company admitting, in the most concrete language available, which story it thinks decides the outcome.
🕒 Published: