A supervised trip measures some things well and leaves the thing that actually decides the business untouched. It tells Wayve whether a Londoner offered a Mustang Mach-E in the app accepts it rather than switching to a conventional car before pickup [6]. It tells Uber whether matching, door unlock and trip start behave when the driver is software [7]. Neither touches the cost of a trip with nobody aboard, because there is somebody aboard every one of them [2].
Run the cost line. Each car carries a Transport for London licensed private-hire driver whose job is to monitor the system [2]. The rider pays ordinary UberX, Comfort or Electric pricing with no premium for the autonomous match [1][6]. So the trip carries a private-hire wage, a Mach-E and whatever the sensing and compute cost, against the revenue of an ordinary minicab ride. The source reporting concedes the point: the fleet is too small to show whether Wayve improves Uber's economics or displaces driver supply [4], and the cost, reliability and remote-support requirements of a driverless service stay open until the safety driver leaves the vehicle [5].
The demand figure has a similar shape. Uber says more than 140,000 Londoners had opted into the relevant ride preference before launch [8]. Against fewer than 20 cars, that is upwards of 7,000 interested riders per vehicle [1], which is a queue rather than a load test, and Uber's own framing is that the number counts expressions of interest and not completed or repeat rides [8].
The launch matters most to the licensing model. Wayve intends to sell its driver to automakers and fleets rather than own a service end to end, which is why Mercedes-Benz, Nissan and Stellantis sit on the cap table alongside Microsoft, Nvidia and Uber [13][14]. The technical claim underneath is that an end-to-end model trained on driving data, without high-definition maps or city-specific rule engineering, adapts across roads, cities and vehicle platforms [9]. For a London result to transfer to a licensee, three conditions would have to hold that London does not test: a different vehicle and sensor platform, a city Wayve has not been training on since 2018 [10], and no licensed supervisor sitting behind the wheel as the backstop. The one metric that would begin to settle it, an intervention rate, is not among the numbers disclosed at launch [3].
The passenger-facing engineering is the good part. Consent is explicit before pickup [6], and the in-car screen renders the vehicle's planned path in 64 languages [7], which is a cheap and effective way to make the machine's intent legible. The screen shows riders where the car means to go, but what the ride cost to provide is a separate question altogether, one it does not answer. If the supervised phase earns its keep, the artefact it produces is an intervention log good enough to put in front of the regulator whose authorization is required before that seat can be emptied [3], which is roughly how Uber's Sarfraz Maredia described the goal to Reuters: credibility with consumers and with government [11].</body_markdown> </invoke>
