Waymo’s playbook for opening a new city has become familiar enough that it reads like a recipe. Map the operational design domain. Run safety drivers for six to nine months. File the permits. Open a waitlist. Charge fares. Quietly remove the geofence one neighborhood at a time. The recipe has worked four times in a row, which is more than any other autonomous vehicle company can say about anything.

The question is whether the recipe survives the move from “open one city per year with a hand-picked launch team” to “open four cities in parallel with people who have never done this before.”

What the four launches actually proved

Phoenix proved a robotaxi service could exist at all, in a city that was designed for cars and has weather that is friendly to cameras nine months of the year. The lesson was technical: the stack works when the operational design domain is generous.

San Francisco proved the stack could survive a hostile operational design domain, fog, hills, cyclists who treat lane lines as suggestions, double parking as a permanent state of affairs. The lesson was that the perception problem and the prediction problem are the same problem at scale, and Waymo had been solving them together for longer than the competition realized.

Los Angeles proved the service could grow. Phoenix and SF were proofs of concept the size of one neighborhood each. LA was the first time Waymo had to run an operation that looked like a real transportation business: a depot footprint, a fleet management team, a rider support function that could handle volume on a Saturday night.

Austin proved the cross-staff playbook. The local team in Austin was not the team that launched LA. The systems they used to operate were the systems LA had written down. That handoff working, or not working, was the gating question for everything that came next.

The thing nobody talks about

Waymo’s safety record is, on the available evidence, exceptional. The company publishes data, the data is good, and the third-party comparisons to human-driven Uber and Lyft are consistently in Waymo’s favor by the metrics that matter.

What is not exceptional is Waymo’s unit economics. The vehicles are expensive. The remote assistance staffing is expensive. The mapping is expensive. The HD lidar is expensive. The mapping team’s maintenance of those maps, every time a new development goes up, is the cost line that nobody outside the company seems to talk about, and it is the cost line that scales linearly with city count rather than with miles driven.

The bet Waymo is making is that the per-mile cost comes down faster than the per-city cost goes up. So far, the public financials don’t tell us whether that bet is winning.

The next city is the hard one

Whichever city Waymo opens next, and the rumored candidates have been Miami, Atlanta, and Washington, DC for the better part of a year, the test isn’t the city. The test is the org chart.

Four cities in parallel is when the playbook stops being a checklist that the founders walked through with each local team, and starts being a piece of operational software that has to run without supervision. That is a different company than the one that exists today.