Why the pipe work stays with people
Plumbing is decided in places software cannot reach. Someone has to crawl under a joist, cut an opening in a wall or floor, and get a wrench onto a fitting that is stuck. Assembling pipe sections, tubing, and fittings with couplings, clamps, screws, and solder is slow, variable work in tight spaces, and no two houses are built the same.
The judgment is physical too. Locating and marking the position of pipe installations, connections, and fixtures means reading an old building, not just a drawing. A joist runs where the plan says it shouldn’t. A vent was added later by someone in a hurry. The fix depends on what you find after the drywall comes off.
There is a second reason this trade holds. Plumbers carry liability. Work has to pass code and inspection, and a licensed person signs for it. Software can look up a code clause in seconds, but it cannot be held responsible for a joint that fails above a finished ceiling. That share of task time that needs a person is printed above: 81%.
What software handles, what it assists, and what you still do
The tasks AI can take on by itself are the desk side of the job. Preparing written work cost estimates, pulling material lists, and checking a job against plumbing codes and regulations are text-and-numbers work, and tools already do a usable version of them. The share of task time in that group is 4%.
The assist group is larger in effect than in size. Reviewing blueprints and building codes to work out the layout goes faster with a model that reads the drawing set and flags conflicts. Inspection cameras with image recognition help read a sewer line and suggest where the blockage sits. Scheduling, dispatch, parts ordering, and customer follow-up all compress. That group’s share is 15%. The overall share of task time AI can handle today is 10, and how coverage is measured explains what counts.
What is left is the trade itself: cutting and threading pipe, soldering and fusing joints, installing and repairing fixtures and appliances, testing a system under pressure, and finding the leak nobody else could find. These are hands, eyes, and access problems. A model can tell you the likely cause. It cannot reach the valve.
What the evidence actually shows
There is no direct test of AI against working plumbers yet. The evidence grade for quality parity here is D, and a D grade means nothing has been measured head to head, so no parity number is given. We will not put a figure on something nobody has run.
The best-known estimate is still an estimate. Frey and Osborne (2013) put the probability of computerization for this occupation at 35%, based on expert judgment about task bottlenecks rather than any trial. That was twelve years before the current generation of models, and it tested nothing in a crawlspace.
What would settle it is specific: a timed field trial on real service calls, where a robot and a journeyman each diagnose and repair the same set of faults in occupied homes, scored on first-time fix rate, code compliance, and damage. Until that exists, the honest answer is uncertainty on the parity question and reasonable confidence on access. The full method is set out in how the scores are built.
The labor market is less ambiguous. BLS counts about 465,840 plumbers, pipefitters, and steamfitters in the United States, with median pay of $63,800, and projects employment growth of 6.8% from 2025 to 2035 (BLS, 2025). That is a trade adding workers, not shedding them.
When this could change
Most likely after 2047 (8 in 10 of our scenarios). You can read what that window does and does not measure on the replacement-year method page.
Two things could pull it earlier. The first is a genuine jump in dexterous humanoid hardware: hands that can thread a fitting in a space built for one arm, and legs that can get there. The second is prefabrication. If more plumbing arrives as factory-built modules, the on-site task shrinks before any robot has to be clever, and the field work shifts toward connection and testing. The guide to humanoid robots in physical jobs covers where that hardware stands.
Two things hold it back. Cost is one: the cost panel above sets the machine range against the human range for the same work, and robot capability of this kind is not cheap to buy or maintain. Licensing and inspection are the other. Permits, local code, and liability are built around a named person, and that structure changes slowly even when the technology does not.
What to do: treat the assist tools as a way to cut estimating and paperwork time, not as a threat to the wrench work.
How to stay needed
Lean into the parts of the job that resist remote work. Diagnosis in messy, older buildings, where the fault is not where the symptom is. Fixture and appliance installation in finished spaces, where damage costs more than the repair. Pressure testing and commissioning, where someone has to stand behind the result.
Two skills pay off. Gas, medical gas, backflow, or steam work raises the licensing bar and the rate with it. Running the business side well matters too, and the estimating tools help more if you understand what they are pricing.
If you are weighing nearby trades, pipelayers, electricians, and heating, air conditioning, and refrigeration mechanics and installers share most of the same constraints. You can also see the wider construction trades workers family, the construction sector page, or the jobs that mostly need a person list. To put this trade next to any other job, use the side-by-side comparison.