Why separators, filters and stills stay with operators
This job is a plant-floor job. You start and shut down separators, filter presses and stills, set flow rates, open and close valves, and watch pressure and temperature as a batch runs. When a filter loads up, someone has to break it down, pull the cloth or cartridge, clean the screen and build it back. Software can tell you a differential pressure is climbing. It cannot carry the media or torque the bolts.
The other half of the day is judgment in a messy environment. Operators draw samples, check clarity and specific gravity, and decide whether a batch goes forward, gets recycled or gets held. They smell a seal starting to go, hear a pump cavitate, and notice a line weeping before any sensor trips. Many plants run older skids that were never fully instrumented, so the real control loop is a person with a wrench and a logbook.
That is why the headline figure here, 84 out of 100 (higher is safer), sits where it does. You can read how that figure is built on the Still needs a human page.
What software runs, what it assists, and what people keep
Start with the work software can run on its own. The task split above puts 0% in that group. It is the paperwork end of the job: logging readings into batch records, and tracking trends in pressure, flow and temperature so a shift report writes itself. These are tasks where the data already exists in a historian and nobody needs to touch a valve.
Next, the assisted group: 6%. Here a model suggests and a person decides. It can flag that a filter cycle is running long against the last fifty batches, or prompt a setpoint change when feed quality shifts. The operator still makes the adjustment, signs the record and lives with the result.
Everything left belongs to people: 94%. That is sampling and testing product clarity, changing and cleaning filter media, dismantling equipment for service, and chasing leaks and plugged lines when a run goes wrong. On the Can AI do it? question, coverage scores 7 out of 100; how coverage is measured explains what that counts.
What has actually been tested
Not much, and that matters. The evidence grade for this job prints D, which is our marker for no direct head-to-head test of AI against a qualified operator in this work. So there is no parity number on this page, and we will not guess one.
What would settle it is specific. A published trial where an automated separation or filtration line runs batches against trained operators on the same feed, scored on yield, off-spec rate, downtime and safety incidents. Or a documented filter-change cell, with a robot handling media in a wet room, measured over months rather than a demo day. Until something like that exists, the honest answer is that the physical side has not been measured against people. Our grading rules are set out in the Is it better than a person? method.
When the picture could change
Most likely after 2046 (8 in 10 of our scenarios). What that window means, and how it is calculated, is explained in the replacement-year method.
Two things could pull it in. New plants are built as closed, fully instrumented skids with automated valves, self-cleaning filters and in-line analyzers, so the sampling and adjusting stop being manual at all. And the robotics panel above puts the hardware this job would need in the mobile robots tier: machines that move through a plant rather than sit bolted to a floor. If those get cheap and reliable in wet, hot, slippery rooms, some filter handling moves.
Two things hold it back. Most of the installed base is old and non-standard, and retrofitting a 30-year-old filter press costs far more than paying someone to run it. And safety rules do not bend: lockout and tagout, confined space entry and permit work all assume a trained human signs off. The cost panel above shows the software side is cheap next to a person, but cheap software does not break down a filter press.
The market pressure shows up in hiring, not in disappearance. The Bureau of Labor Statistics counts about 60,100 of these jobs in the United States, with a projected 5.7% decline over 2025 to 2035 and median pay of $51,610 (BLS, 2025). That is fewer openings and fewer entry-level slots, spread over a decade. You can see jobs on a similar track on our list of jobs expected to shrink.
What to do: get named on the plant’s instrumentation and controls work, because that is where the remaining headcount concentrates.
How to stay needed in the filter and still room
Lean into the tasks that stay on the human side of the split. First, sampling and quality calls: being the person who can read a sample and defend a hold decision. Second, teardown and rebuild of filters, columns and stills, including the dirty jobs nobody schedules. Third, troubleshooting a bad run end to end, from feed quality to seal failure.
Two skills pay here. Learn the control system you work on well enough to tune alarms and read trends, not just acknowledge them. And learn to write a clean deviation report, because when a batch goes off-spec, the record is what the plant defends.
If you are thinking about a sideways move, the closest work sits nearby: chemical equipment operators and tenders, mixing and blending machine setters, operators and tenders, and crushing, grinding and polishing machine setters, operators and tenders. The wider other production occupations family and the manufacturing sector page show how those scores stack up, and you can put any two jobs side by side on the compare tool. Our full scoring approach is at the methodology page.