Why the cab job has not gone away
Driverless metro trains already exist. Lines in Vancouver, Copenhagen, Dubai and Honolulu run with no operator in the cab. So when people ask will AI replace streetcar operators, the answer starts with the track and the signal system, not with the software.
Driving a train on a fixed guideway is the easy part. Stops are in the same place every time, speeds are set, and the route never changes. The harder parts of the shift are the ones that happen around the train. Operators watch the platform edge before closing doors and starting away. They take manual control when signals fail or a car blocks the crossing. They handle sick passengers, fights, fire alarms and evacuations onto a live right of way. They answer riders who are lost, late or angry. A streetcar operator also shares the lane with traffic, cyclists and people stepping off the curb.
Scale matters too. This is a small, well-paid occupation: about 10,200 people, with median pay of $86,380 (BLS, 2025). Employment is projected to change by 3.7% between 2025 and 2035 (BLS, 2025 projections). Replacing those operators is not a software purchase. It means resignaling a line, often adding platform screen doors, and proving the whole system to a safety regulator. You can read how we turn facts like these into scores on our scoring methodology page.
What is automated, what is assisted, and what stays with the operator
Automatic train operation is old technology. Computers already regulate speed between stations, hold a steady headway and stop the train at the mark more smoothly than most people can. Equipment readings and fault logs are captured without anyone writing them down. In our split, the share of task time in this group is 0%.
Assistance is where most of the current gain sits. Control centers use software to adjust spacing when a train runs late, flag a door or brake fault before it becomes a delay, and push announcements from live service data. Camera systems can watch platform crowding and alert staff. The share of task time in the assisted group is 28%. None of this removes the person; it changes what the person pays attention to. Our coverage method explains how task time is counted.
Then there is the work that still needs a person on board or on the platform. Judging whether it is safe to close the doors on a crowded platform is a live decision, not a rule. So is taking the train forward by hand when the signal system drops into degraded mode, clearing passengers through a tunnel, or reading a street-running hazard that no track circuit can see. The share of task time left to people is 72%.
What the evidence can and cannot tell us
There is no published head-to-head test of an AI system against a US subway or streetcar operator doing the full job. Our evidence grade for this occupation is D, and a grade of D means the comparison has not been measured, so we give no parity number at all.
What would settle it is specific. Incident and injury rates on automated lines compared with staffed lines on similar track. Performance when the signal system fails and the train has to be driven by hand. Evacuation outcomes with and without a trained person on board. Door-closing incidents on crowded platforms. Until that kind of record is published and audited, claims in either direction are opinion. Our quality parity method sets out the bar.
When this could realistically change
Most likely after 2044 (8 in 10 of our scenarios). The replacement year method explains what that window is based on.
Two things could pull it closer. New lines and extensions are increasingly designed for automated operation from day one, which skips the retrofit problem. And resignaling programs on older systems, usually with platform screen doors, make driverless running possible on track that already exists.
Two things hold it back. Cost and time: a full resignaling runs for years on a working railroad, and the equipment bill dwarfs the cost of running AI tools. And geometry: street-running streetcars mix with cars and pedestrians, which is a far harder environment than a sealed tunnel. Labor agreements and state safety rules add more friction, and many agencies that automate still keep staff on board for incidents and customer service.
What to do: if your agency announces a resignaling or CBTC program, ask early about the training path into control-center and supervisory roles, because those seats are filled before the new system goes live.
How to stay needed in transit operations
Lean into the parts of the job no control system owns. First, incident handling: evacuations, medical events, fire alarms and police holds. Second, degraded-mode operation, where you move a train safely with the automation switched off. Third, platform and street judgment, where you read people rather than sensors.
Two skills travel well from here. One is rail traffic control and dispatching, since automated lines need more people watching the whole system, not fewer. The other is safety and incident command, including the paperwork and the testimony that follows an event. Both move you up the chain rather than out of it.
Nearby work is worth a look. The cab skills carry over to locomotive engineers, while the passenger and incident side lines up with railroad conductors and yardmasters and with transit and intercity bus drivers. You can see all of these together on the rail transportation workers family page, or in the wider transportation and warehousing sector.
Want to test a move? Put two of these jobs side by side with our job comparison tool, or scan the jobs most exposed to AI list to see where this kind of work sits against the rest of the labor market. The headline figure for this occupation is 79 out of 100 (higher is safer).