Olivia Solon rode one for Bloomberg on a wet morning and described the trip as an autonomous ride with hands still hovering over the steering wheel [18]. The hovering is a licensing artefact. Between the day the Automated Passenger Services route opened and the day the fare meter started, 111 days passed [1], and the first mover picked the door that ships sooner [7].
That choice is defensible on its own terms. Running supervised cars with real passengers builds exactly the operational record an APS application will eventually need [7], which is a better use of a year than waiting in a queue. The price is that every kilometre is staffed.
Then there is the number the launch offered as proof of appetite. Wayve says more than 140,000 Londoners have opted in through Uber's trip preferences [11]. The fleet is 15 cars, covering everywhere in London except the airports [12]. That works out to roughly 9,300 opted-in riders per vehicle [2]. Here is what a trip-preferences toggle actually records: a person said yes once, in a settings menu, to a car type they may never be matched with. Here is what teams tell themselves it records: demand. It does not tell you whether anyone rode twice, how deep into the week the novelty holds, or how often the licensed driver's hands came down onto the wheel.
Separate the thing being pitched from the thing being done. The pitch is an AI Driver that learns from experience instead of depending on HD maps and hand-coded rules, marketed as AV2.0 [13], trained on London roads since 2018 [14], funded by roughly $1.5bn in a Series D that valued the company near $8.6bn on the promise of generalising across cities rather than surveying each one [15]. Chief executive Alex Kendall framed the launch as the start of a global rollout with Uber [19]. The thing being done is a small supervised fleet on private hire plates, where a licensed human is legally answerable for what the software decides [4].
The grid worth drawing, if you are shipping the supervised version of anything: does the legal or contractual route available to you today require a human in the loop, and does your value proposition depend on removing that human. Wayve sits in the required-and-depends box, which is the expensive corner, and the only exit is that supervised operation produces the evidence a permit demands. Supervision keeps the labour cost and shifts the liability onto the person in the seat; what it buys is time on real roads. Waymo, which has signalled its London cars may carry no driver from the start [16] and already runs fully driverless in more than a dozen American cities [17], is betting the other way in the same city.
What settles the comparison is the share of trips the software finishes without help, and that is a permit-file number rather than an app-menu one.
