Why the work stays inside the plant
Stationary engineers and boiler operators keep steam, heating, cooling and power systems running in hospitals, factories, schools and office towers. The job runs as a loop: read gauges, test boiler water and dose chemicals, adjust valves and dampers, chase a noise or a smell, then sign the log. Software can read the gauges faster than anyone. It cannot repack a leaking valve, pull a burner for cleaning, or stand in front of an inspector. That is the simple reason most task time here stays with people: share of task time that needs a human is part of the headline score, and for this job it sits at 83%.
Two task groups carry that weight. The first is physical maintenance: lubricating pumps and compressors, replacing gaskets and bearings, cleaning tubes, tightening packing. The second is response. When an alarm says one thing and the equipment says another, somebody has to walk the floor, decide whether to reduce load or shut down, and own that call. Licensing reinforces it. In most cities a boiler of any size needs a licensed operator attached to it, and the license belongs to a person, not a control system.
The market is small and stable rather than booming. BLS counts about 28,250 of these jobs in the United States, with median pay of $78,620, and projects employment growth of roughly 3.1% from 2025 to 2035 (BLS, 2025). Modest growth, high responsibility, and a licensing gate that keeps the supply of qualified operators tight.
What AI runs, what it assists, and what it leaves alone
Monitoring is the part machines already do well. Building automation and plant control systems track pressure, temperature, flow and water level around the clock, record readings, and raise alarms without anyone watching a panel. Report writing and log keeping sit in the same bucket. Across this job, the task time AI could handle today is 14 out of 100, and the share of affected work it could run without a person is 0%. The coverage method explains what counts as handled.
Assistance is where the bigger change is landing. Predictive maintenance tools flag a pump or fan drifting toward failure before it trips, and load-scheduling software tunes setpoints to cut fuel use across a shift. Fault diagnosis is getting the same treatment: a model can rank likely causes of a high stack temperature while the operator walks to the boiler. The share of affected task time that looks like help rather than handover is 17%.
What stays with the operator is everything that needs hands, eyes and a signature. Starting up and shutting down equipment safely. Repairing and adjusting valves, burners, traps and controls. Testing water chemistry and correcting it. Working with inspectors and contractors during an outage. A remote system can tell you a reading is wrong; it cannot tell you the sensor itself is wrong, which is a judgment experienced operators make weekly.
The evidence: no head-to-head test yet
The quality grade for this job is D, and a D grade means one plain thing: nobody has published a direct test of AI against a licensed operator doing this work. There is no parity number for that reason, and we will not invent one. Benchmarks that do exist cover office and coding tasks, not steam plants.
What would settle it is specific. A published trial where an automated plant handled startup, shutdown, water chemistry and an unplanned trip over months with no operator on site. Audited staffing data from buildings that moved to remote operating centers, showing how many licensed hours actually disappeared. Robot trial results on real maintenance tasks, not staged demos. Until something like that exists, the honest read is task erosion in monitoring and record keeping, with the physical and licensed core intact. Our scoring method grades every job this way, so a thin evidence base shows up as a grade rather than a confident claim.
When it could change
Most likely after 2046 (8 in 10 of our scenarios). The replacement-year method sets out how that window is built.
Two things could pull it earlier. Software is cheap next to a staffed shift, so the business case for centralized, remote monitoring of many buildings is already strong, and it keeps improving. And general-purpose robots are the live variable: most of the hands-on work here would need the dexterous humanoid tier, so real progress there changes the math quickly.
Two things hold it back. The physical share of the task mix is large, and plumbing, burners and rotating equipment are unforgiving of clumsy hands. Licensing and code are the second brake: jurisdictions tie boiler operation to a named, certified person, and that rule changes slowly because the failure mode is an explosion or a hospital losing heat. Old plants add friction too, since many boilers in service predate any usable digital interface.
What to do: get the controls and analytics work onto your resume now, while the monitoring tasks are still shifting rather than settled.
How to stay needed
Lean into the work that needs a licensed body on site. First, emergency response and shutdown decisions, including documenting what you did and why. Second, mechanical repair and adjustment across boilers, pumps, chillers and valves, which is where your rate is hardest to replace. Third, inspection and compliance work with city inspectors and insurers, where the license and the relationship both matter.
Two skills raise your floor. Building automation and SCADA fluency, so you own the system that watches the plant instead of being replaced by its alerts. And water treatment plus combustion tuning, because efficiency and boiler life depend on judgment that software can only suggest.
If you are weighing a move, the closest work sits nearby: power plant operators, water and wastewater treatment plant operators, and gas plant operators. All three share the control-room-plus-hands pattern. You can see how they stack up on the plant and system operators family page, or put any two side by side with the job comparison tool. For the wider picture, the utilities sector page covers employers in this space, and our guide to humanoid robots and physical jobs explains why the robotics timeline matters more here than any chatbot does.