$110bn. That’s Waymo’s valuation after finalising a $16bn funding round, and it’s the number I keep coming back to when people ask me why their bot project stalled at 80% done.
Because the news cycle around robotaxis right now has almost nothing to do with perception stacks or path planning. Waymo has doubled its lobbying spend. Uber has pledged $10bn to win the robotaxi race. Waymo is reportedly exploring a split with Uber as tensions between them deepen. Volkswagen is pushing into robotaxis to challenge Waymo and Tesla. Five headlines, and the closest any of them gets to engineering is the word “self-driving.”
I build bots for a living. Conversational agents, task runners, automation that talks to APIs and occasionally to humans. And the pattern in this robotaxi fight is the most useful thing I’ve read all month, because it’s the same wall every bot builder hits, just scaled up to billions.
The demo is not the deployment
Every bot I’ve shipped has had two distinct phases. Phase one is fun: get the model responding, wire up the tools, watch it do something impressive in a terminal. Phase two is the part nobody blogs about. Rate limits. Permissions. Who owns the data. Which team’s approval you need. What happens when it’s wrong in front of a customer.
Waymo doubling its lobbying budget is phase two at civilizational scale. The cars work well enough that the constraint has moved. When your bottleneck becomes regulators and partnership terms rather than model accuracy, that’s not a failure. That’s a signal you’ve cleared the technical bar and hit the institutional one.
The tell is the Uber situation. Waymo exploring a split from a distribution partner while that same partner commits $10bn to compete directly — that’s a fight over who owns the customer relationship, not who has better sensors. If you’ve ever built a bot on top of someone else’s platform, you know exactly where this is going.
Distribution is the actual moat
Here’s what I’ve learned building on other people’s rails. The platform that owns the user will eventually want what you’re building. Not out of malice, just gravity. You prove the demand, you make the integration look easy, and the incentive to keep paying you erodes.
Uber had Waymo vehicles routing through its app. Uber now has $10bn earmarked for this race. Waymo has $16bn and is looking at the exit. Both parties learned the same lesson from the same partnership.
Practical version for anyone building bots:
- If your bot only reaches users through one channel, that channel is your single point of failure, and it’s not a technical one.
- Design your core logic so the interface is swappable. Slack today, WhatsApp tomorrow, your own web app when the platform changes terms.
- Own your evaluation data. Own your conversation logs, within whatever privacy rules apply. If a partner holds those, they hold your ability to improve.
- Read the terms of service like they’re a threat model, because they are.
Volkswagen and the cost of being early
Volkswagen entering to challenge Waymo and Tesla is the other half of the story. A legacy manufacturer moving in tells you the category is now credible enough to justify a large industrial bet. Which means the pioneers spent years and enormous capital establishing that a market exists, and now competitors get to skip that part.
Same thing happens in bot land, faster. Someone builds a clever agent workflow, publishes the approach, and three months later it’s a feature in a product with a hundred million users. The lesson isn’t don’t build. It’s know which part of your work is defensible. Usually not the prompt. Sometimes the data. Often the boring integration nobody else wants to maintain.
What I’m taking into my next build
I’m not going to pretend a $16bn round has direct lessons for a side project. But the shape of the problem transfers.
Budget for phase two from the start. If you’re building anything that touches regulated data, real money, or people’s safety, the approvals and the paperwork are part of the build, not an afterthought. Waymo apparently allocated enough to double a lobbying line item. Your version might be a week of security review or a compliance doc, but it’s the same category of work.
Assume your distribution partner is also a future competitor. Build accordingly. Abstract the interface layer, keep your own metrics, and never let a single integration become load-bearing for your whole product.
And treat model quality as table stakes rather than strategy. The robotaxi fight moved to lobbying budgets and partnership disputes precisely because the driving got good enough. Your bot will get there faster than you expect, and then the interesting problems start.
The engineering was always the tractable part. Everything after it is where products live or die.
🕒 Published: