Why the plant still needs an operator
Water and wastewater plants already run on automation. Sensors track turbidity, pH, flow and chlorine residual all day, and a control system can hold a plant steady for long stretches without anyone touching a dial. That is not new, and it is not the same as removing the operator. Someone still has to walk the basins, listen to a pump that sounds wrong, and decide whether an odd reading is a real shift in the raw water or a fouled probe.
Two tasks make the point. Collecting and testing samples is physical work in wet, confined, chemical-heavy spaces. Adjusting chemical feed as source water changes after a storm, a spill or a seasonal turnover is a judgment call with legal weight behind it, because the plant has a permit and a named certified operator on duty. Add the repair side: clearing a clogged filter, backwashing, resetting a failed pump, closing a valve by hand when the actuator quits.
The robotics picture on this page shows why software alone does not finish the job. About 72.7% of the work has a physical component, and the robot class it would need is mobile robots that can move around plant grounds and into tanks and galleries. That hardware exists in other settings. It is not standard in municipal treatment plants, and the environment is unkind to it.
What software does, what it helps with, and what stays with you
The tasks marked as work AI can handle on their own sit on the paperwork side: pulling readings together, building routine logs and drafting the regular reports that regulators expect. That share is 0% of task time. Record-keeping is the part of the day most likely to shrink first.
A larger block of the job is shared work, where the software suggests and the operator decides. That share is 16%. Think of dosing recommendations from a model that has seen years of plant data, or alarm triage that ranks which of forty notifications actually matters. The operator still signs off, and still owns the result. Overall task coverage sits at 12 out of 100, measured the way the coverage score describes.
Everything else needs a person on site: 84% of task time. That is sampling and bench testing, hands-on maintenance, and the response when a lift station floods or a digester goes sour at 3 a.m. No model can open an access hatch.
What has actually been tested
Honest answer: not much, for this job. The evidence grade here is D, which means no study has put an AI system head to head against licensed operators in a working plant and published the result. So this page gives no quality-parity number, and you should treat anyone who offers one with care.
What would settle it is specific. A published trial of model-driven chemical dosing against certified operators across changing raw-water conditions, scored on permit compliance, upset events and chemical use. Alongside it, an audited comparison of alarm handling during real failures, not simulations. Until something like that exists, the fair reading is that automation supports the control room and the lab bench without a measured replacement of the operator’s judgment. The quality-parity method explains how a graded score would be built if that work appeared.
When this could change
Most likely after 2046 (8 in 10 of our scenarios). How that range is built is set out in the replacement-year method.
Two things could pull it earlier. The first is remote consolidation: one staffed control room watching several small plants, with fewer people on each site. The second is cost and headcount pressure. The Bureau of Labor Statistics counts 128,490 of these jobs with median pay of $60,020, and projects employment falling 5.7% between 2025 and 2035 (BLS, 2025). Shrinking payrolls change the job before any robot does.
Two things hold it back. State licensing and permit rules tie accountability to a certified human operator on shift, and those rules move slowly. And the physical environment is brutal: corrosive air, standing water, confined spaces and equipment that is decades old, which is a hard setting for mobile robots and an expensive one for retrofits. Municipal capital budgets run on long cycles, so plant upgrades land over years, not quarters.
What to do: keep your certification current and move up a grade when you can, because the license is what the regulator recognizes, not the control system.
How to stay needed
Lean into the parts of the shift that have no digital substitute. First, upset response: diagnosing why effluent quality slipped and fixing it before a violation. Second, hands-on maintenance and repair of pumps, valves, blowers and filters. Third, sampling and bench work, including knowing when an instrument is lying to you.
Two skills raise your floor. One is instrumentation and controls: calibration, loop checks, and enough SCADA knowledge to spot a bad tag or a drifting sensor. The other is reading data critically, so when a model recommends a dose change you can say why you accept or override it. That record of overrides is also what keeps a human in the loop on paper.
If you are weighing nearby work, these jobs share the same plant-and-system-operator family: chemical plant and system operators, power plant operators and stationary engineers and boiler operators. You can see all of them on the plant and system operators family page, and the wider picture on the utilities sector page. To weigh two roles side by side, use the job comparison tool, or scan the list of jobs that mostly need a person. Our scoring approach is documented in full on the methodology page.