Why a person stays at the controls
A freight train can stretch more than a mile and weigh thousands of tons. The engineer reads signals and track conditions, works the throttle and the air brakes, and keeps the train inside the railroad’s operating rules and federal regulations. Stopping distance is measured in thousands of feet, so most of the skill is in judgment made minutes before anything happens.
The rest of the job sits outside the cab. Engineers inspect the locomotive before and after a run, listen and look for mechanical defects, report what they find, and handle the unplanned: a grade-crossing incident, a trespasser, ice on the rail, a radio call from the dispatcher telling them the plan just changed. The question behind this page — will AI replace locomotive engineers — comes down to how much of that work is hands-on, local and rule-bound. Most of it is.
Scale matters too. About 33,470 people worked as locomotive engineers in the United States, with median pay near $81,410 a year, and projected employment change of roughly 0.5% from 2025 to 2035 (BLS, 2025). That is a small, stable occupation sitting on top of a very large installed base of track, signals and rolling stock. Changing how trains are driven means changing that hardware, not just the software. You can see how we weigh all of this in our scoring method.
What software already does, what it assists, and what it leaves to the engineer
The paperwork side of railroading has been digital for years. Run data, event recorder output and defect reports move into railroad systems with little human typing, and dispatch software proposes meets, passes and train sequencing across a territory. Our task split puts 0% of task time in the group machines can handle without a person in the loop, and our coverage measure explains how that share is counted.
Assistance is where the real change has happened in the cab. Positive train control enforces speed restrictions and stop signals behind the engineer. Energy-management systems recommend throttle and braking settings to save fuel and keep slack under control. Wayside and onboard sensors flag hot bearings and dragging equipment. None of that drives the train on its own; all of it changes what the engineer watches. Assisted work comes to 4% of task time.
What is left is the core of the job: handling the train when conditions degrade, confirming signals and restrictions by eye, judging a crossing situation in seconds, and coordinating by radio with the conductor and the dispatcher. Work that still needs a person accounts for 96% of task time here. A large part of it is physical and happens off the seat, which is why the robotics panel above puts this job in a hardware tier that does not exist as a product today.
What has actually been tested
Our evidence grade for quality parity is D. In plain terms: there is no published head-to-head test of an AI system against a qualified locomotive engineer on US mainline work, so we give no parity number. Claiming one would mean inventing it.
That is not the same as saying automation does not work on rails. Fully automatic operation runs today on sealed metro and airport-shuttle lines, where the right of way is fenced, stations are controlled, and the route is fixed. Mainline freight is a different problem: shared track, public grade crossings, variable train makeup, weather, and crews who also inspect and troubleshoot. Automation that is proven in one setting does not transfer to the other without new evidence.
What would settle the question is a published, independent comparison on mixed-traffic track: train handling across grades and in poor adhesion, response to unplanned obstructions and crossing incidents, and defect detection on inspection, measured against experienced engineers over many runs. Until something like that exists, this page stays a grade with no number, and we explain why in our quality parity method.
When automated trains could change this job
Most likely after 2044 (8 in 10 of our scenarios). What the window measures, and how it is built, is set out in the replacement-year method rather than restated here.
Two things could pull the date earlier. The first is corridor design: dedicated or sealed freight routes with grade separation remove the hardest part of the problem. The second is remote supervision, where one operator oversees several trains that run under automatic control and a person only steps in by exception. Both have precedent in mining and metro operations.
Two things push it later. Grade crossings and public right-of-way mean a single bad judgment carries community-scale consequences, which keeps regulators cautious and the burden of proof high. And the fleet itself is the brake: retrofitting locomotives, signals and inspection practice across a national network takes capital and years, no matter how capable the software gets. Crew-size rules and labor agreements sit on top of that.
Good to know: the change most engineers will feel first is not an empty cab, it is more enforcement and advisory systems in the one they already sit in.
How to stay needed
Lean into the parts of the job that software cannot carry. Train handling in degraded conditions — wet rail, heavy tonnage, long grades — is still judgment built from miles. Emergency response and grade-crossing decisions are yours in the moment, and the way you report them shapes the record. Pre-trip inspection and defect reporting keep you in the part of the work that needs hands, ears and a walk around the locomotive.
Two skills compound. The first is working well with automation: knowing what positive train control and energy-management systems do, when their advice is wrong, and how to take over cleanly. The second is incident communication — clear radio work, clear write-ups, clear handoffs — which is what railroads lean on when something goes wrong.
Nearby work is worth reading alongside this page: Railroad Conductors and Yardmasters, Rail Yard Engineers, Dinkey Operators, and Hostlers and Railroad Brake, Signal, and Switch Operators and Locomotive Firers. For the wider picture, see the rail transportation workers family and the transportation and warehousing sector.
You can put this job next to another on our side-by-side compare tool, or read why physical work moves slowly in our guide to humanoid robots and physical jobs.