Picture a conference room in Washington, D.C. Someone from a rideshare company is handing a state legislator a draft bill. The draft says human drivers must handle 85% of trips. That’s the New Jersey version, according to reporting on the Uber-Waymo rift, and it’s the kind of detail that tells you more about the autonomous vehicle business than any demo video ever will.
I review tools for a living. I install things, break them, and write down whether they held up. So my instinct when a story like “Waymo doubles spending on lobbying in robotaxi battle with Uber” crosses my feed is to ask the question I ask about every product: what problem is this actually solving?
The part nobody puts in the spec sheet
Here’s what we know from the verified reporting. Alphabet-owned Waymo has doubled its lobbying expenditures. The goal is influencing U.S. regulations around robotaxi operations. Uber is on the other side of that fight, and in at least one case — that New Jersey legislation — Uber lobbyists have worked alongside unions to push rules requiring human drivers for the vast majority of trips. There’s also a D.C. bill in the mix.
That’s the factual floor. It’s thin, and I’m not going to pad it with invented numbers. But the shape of it is clear enough to reason about.
When I evaluate an AI toolkit, I look at things like documentation quality, error handling, how gracefully it fails, whether the maintainers respond to issues. What I’ve never had a column for is “how much does the vendor spend convincing lawmakers this category should be legal.” For robotaxis, that column might matter more than the model architecture.
Why this is a tooling story, not just a policy story
A lot of AI products live or die on technical merit. A code assistant either autocompletes usefully or it doesn’t. You can test it in an afternoon. Autonomous driving is different, and the difference is instructive for anyone thinking about where AI actually gets deployed.
The self-driving stack could be excellent and still be illegal in your city. Or it could be mediocre and still get a permit because the right bill passed. Technical capability and market access have come apart. That’s a real thing to understand about this category, and it’s not unique to cars — it’s what happens to any AI system that touches a regulated industry.
So when Waymo doubles its lobbying spend, that’s a company reading the situation correctly. The constraint isn’t the perception model anymore. The constraint is a state assembly.
The 85% number is the interesting one
Of everything in the verified facts, the New Jersey provision sticks with me most. A rule that human drivers must perform 85% of trips isn’t a safety standard. It’s not a testing requirement or a data-reporting mandate. It’s a quota that caps how much of the market automation can take, regardless of how well the automation works.
You can argue for that policy on labor grounds and it’s a legitimate argument. Driving jobs are real jobs and people depend on them. But as a piece of technical regulation, it’s blunt. It doesn’t ask whether the system is safe. It asks how many trips it’s allowed to have.
And that’s the tell. Both companies are fighting over allocation, not validation. Uber has a large human-driver network and benefits from rules that require human drivers. Waymo doesn’t, and benefits from rules that don’t. Each side’s regulatory preference maps neatly onto its existing business. Neither position is principled; both are rational.
What I’d tell someone evaluating this space
- Treat regulatory posture as a product attribute. If you’re building on top of an autonomy platform, the vendor’s ability to operate in your target market is a dependency, same as any API.
- Be skeptical of coverage-area claims that don’t mention permits. Geographic availability in this category is a legal artifact more than a technical one.
- Watch state-level activity, not just federal. The reported fights are in D.C. and New Jersey. The rules that decide this will be written in a lot of places at once.
- Don’t read lobbying spend as a proxy for product quality. It’s a signal about where a company sees its blockers, nothing more.
The honest verdict
I can’t tell you from this reporting who has the better robotaxi. That’s not what the story is about, and pretending otherwise would be filling in blanks I don’t have.
What I can tell you is that the fight has moved somewhere most AI tool evaluation never looks. We got good at benchmarking models. We’re still bad at reading the legal surface a model has to operate on. Waymo’s doubled budget is a reminder that for some categories, that surface is the whole game — and no amount of solid engineering routes around a bill that says you’re capped at 15% of trips.
🕒 Published: