Better AI will not decide who wins the robotaxi race. That may sound like heresy coming from someone who builds bots for a living, but the latest news out of the autonomous vehicle world makes the case for me. According to the Financial Times, Waymo has doubled its spending on lobbying in its robotaxi battle with Uber. Not doubled its compute budget. Not doubled its simulation fleet. Doubled its lobbying.
Meanwhile, Uber has pledged $10bn to win the robotaxi race, and the FT frames the company as fighting to stay in that race at all. Put those two headlines side by side and a picture emerges that most engineering blogs are too polite to draw. The decisive layer of the autonomy stack is not perception, planning, or control. It is policy.
Regulation as a System Dependency
As a bot builder, I think in terms of dependencies. Your bot is only as good as the weakest thing it relies on, whether that is an API rate limit, a flaky model endpoint, or a webhook that times out. Robotaxi companies have a dependency most of us never model in our architecture diagrams. They need permission to exist, city by city, state by state.
That permission layer behaves like any brittle external service. It can change without warning. It has no SLA. And you cannot patch it with code. You patch it with lobbyists, which is exactly what Waymo’s doubled spending suggests it understands better than anyone.
This is the part of “shipping AI” that tutorials never cover. We teach people to fine-tune models and wire up agent loops. Nobody teaches them that the deployment environment for physical-world AI includes legislatures, city councils, and public opinion. Waymo’s budget shift is an admission that those environments require as much engineering as the driving stack itself, just a different kind.
What $10bn Actually Buys
Uber’s $10bn pledge is the other half of the story. That number is meant to signal seriousness, and it does. But money at that scale is not primarily buying algorithms. It is buying position. Fleet deals, partnerships, market presence, and yes, influence. When the FT describes Uber as fighting to stay in the race, the fight is not happening in a research lab. It is happening in the market and in the rooms where rules get written.
For those of us who build software, this is a useful correction to a comfortable myth. The myth says the best technology wins. The reality, at least in autonomy, says the best technology that is allowed to operate wins. Those are very different competitions, and Waymo appears to be entering both.
Lessons for the Rest of Us
Most of us are not deploying two-ton robots on public roads. But the same dynamic shows up at smaller scale, and it is worth internalizing early.
- Your bot’s real constraints are often external. Platform policies, API terms of service, and content rules shape what you can ship more than your model choice does. Treat them as first-class architecture concerns, not afterthoughts.
- Budget for the non-code work. Waymo doubling lobbying spend is the enterprise version of a solo developer spending a week on compliance paperwork instead of features. Painful, unglamorous, and often the difference between shipping and not.
- Position beats polish. Uber’s $10bn is a bet that being present in the market matters as much as having the best stack. In the bot world, being first to a platform or integration often outweighs having marginally better prompts.
The Uncomfortable Takeaway
I would love to tell you that the robotaxi winner will be crowned by disengagement rates and elegant sensor fusion. The spending patterns say otherwise. Waymo is arming up on the policy front. Uber is arming up on the capital front. Both companies have concluded that the driving problem, hard as it is, is not the only problem, and maybe not even the main one right now.
For builders, the lesson is not cynicism. It is completeness. A deployed system includes every layer it depends on, and some of those layers wear suits. The teams that map all of them, from the neural network to the statehouse, are the ones that get to keep operating. Everyone else builds impressive demos that never leave the garage.
🕒 Published: