Why the fare still comes with a person
Driving is the visible part of the job. It is not the whole job. A taxi driver loads a suitcase into a trunk, finds a door number on a dark street, and decides whether a curb is safe to stop at with a bus behind. They help an elderly passenger in and out of the back seat. They judge a drunk rider at 2 a.m. and a parent with a car seat at 7 a.m. Software can plan a route. It cannot lift a wheelchair frame or calm a nervous passenger who is late for a flight.
The hardware matters more here than the model. Most of this occupation is physical work in an open world, and the robotics tier shown above is mobile robots, not a fixed arm on a factory line. That means a vehicle, sensors, mapping, cleaning, charging, depots and remote support staff before a single ride is sold. Each new city is its own project, with its own weather, permits and street layout.
Demand is not falling either. Federal data puts employment for this occupation in the tens of thousands, with median pay of $42,100 and projected growth of 11.5% between 2025 and 2035 (BLS, 2025). The pressure that shows up first is usually on fares and hours in the cities where driverless fleets run, not on the trade as a whole. Wider conditions for the ride sector are covered on the transportation and warehousing sector page.
Which parts of a shift software already runs
Start with the work AI does on its own. The share of task time in that group comes out at 6%. It covers the things a dispatcher and a meter used to do: matching a request to a nearby car, picking and re-picking the route around traffic, and calculating and taking the fare without cash changing hands.
Next, the work AI helps with sits at 25% of task time. Turn-by-turn guidance to an unfamiliar address is one. Trip records and shift logs are another: the app keeps the mileage, times and receipts that drivers once wrote down. The driver still does the task, faster and with fewer mistakes.
The rest, 69% of task time, stays with people. Helping passengers and their luggage in and out of the car is in that group. So is handling what goes wrong in the vehicle: a sick rider, a dispute over a route, a child seat that will not clip, a street blocked by a delivery truck. How that split is measured is set out on the coverage method page.
What the testing does and does not show
Evidence quality for this job is graded D on our scale. That is the lowest grade, and it means something specific: there is no published head-to-head test of AI against working taxi drivers across the full job. Driverless fleets publish safety miles and disengagement data. That is not the same as a like-for-like comparison of a complete shift, including the pickups, the assistance and the awkward curbs.
So no parity number is given here. A number would imply a measurement that does not exist yet. What would settle it is a public, audited study of matched routes in the same city, same hours and same weather, counting completed trips, assisted boardings, cancellations and incidents for both a human-driven cab and a driverless one. Until something like that is published, the honest answer is that the quality question is open. The grading scale is explained in the scoring methodology.
Good to know: a high safety record in one mapped city does not transfer automatically to a snowy, unmapped one.
When this could change
Most likely after 2043 (8 in 10 of our scenarios). What the range measures is described on the replacement-year method page.
Two things could pull that earlier. One is city-by-city permits: once a regulator approves driverless passenger service, fleets scale inside that market quickly. The other is unit cost. The per-mile software cost of an autonomous system is small next to a driver’s annual pay, so operators have a strong reason to keep buying down the hardware and remote-support bill.
Two things push the other way. First, the physical share of the job: boarding help, luggage and vehicle condition are not solved by better driving software, and cleaning and servicing a shared car has its own labor cost. Second, geography and weather. Snow, flooding, roadworks and unmapped rural routes remain hard, and the fleets stay in a short list of friendly cities. The guide to robots in physical jobs goes into why hardware timelines slip, and jobs facing the most exposure shows where driving sits against other work.
How to stay needed behind the wheel
Lean into the parts of the job that are not driving. Three are worth naming. Passenger assistance, especially wheelchair-accessible and medical transport work, where a license and a trained pair of hands are the service. Dealing with problems in real time: reroutes, lost property, disputes, a rider who needs a different drop-off. And the local knowledge a stranger pays for: which hotel entrance actually works, which gate at the airport, which street floods.
Two skills raise your floor. Accessibility and passenger-handling certification, which opens contract work that fleets do not bid for. And account relationships: schools, clinics, hotels and corporate accounts book people they trust, not apps. Named work is harder to undercut than street hails.
Close trades are worth a look if you want to move sideways. Compare the task mixes for Shuttle Drivers and Chauffeurs, Bus Drivers, Transit and Intercity and Light Truck Drivers. The wider motor vehicle operators family lists the rest. You can also put two of them side by side on the compare tool before you spend money on a new license.