Why this trade stays in human hands
Sheet metal work is physical, varied and site-specific. A worker measures and marks stock, cuts and bends it to a drawing, then fits and fastens the finished piece into a building that never matches the plan exactly. Ducts have to clear beams, pipes and wiring that moved after the drawings were signed off. That kind of judgment happens with a tape measure in one hand, on a ladder, in a ceiling cavity.
Software is already good at the paper side. Nesting a cut layout, estimating material, driving a CNC press brake or plasma table: these are solved problems in fabrication shops. What has not been solved is the install. Hanging and sealing a duct run, flashing a roof penetration, or re-cutting a transition on the spot calls for hands, balance and a read of the space in front of you.
The share of task time our model leaves with a person is 84%, which reflects how much of the day is spent fitting, fastening and fixing rather than drawing or calculating. Our robotics panel places this work in the Mobile robots tier, meaning the machine would have to move through an unfinished building, not sit bolted to a shop floor.
What AI does, helps with, and leaves to people
Shop-floor computing handles a real slice of the job already. Layout and nesting of parts, cut programs for CNC equipment, and material takeoffs from a model are routine in a modern fabrication shop. The share of task time our model puts in the fully automated group is 0%.
A larger set of tasks is assisted rather than taken over. Reading and interpreting blueprints and specifications is faster with a model viewer and a search tool. Estimating labor and material, documenting installs, and checking a fabricated piece against the drawing all sit here. The assisted share is 16%. In each case a person still signs off and still carries the piece to the hanger.
Everything else is work with tools in hand. Shaping and bending metal by hand or with shears and brakes, welding, riveting and soldering seams, installing duct systems and metal roofing, and troubleshooting a fit that is out by half an inch. Those tasks also explain our Coverage figure of 9 out of 100. Coverage asks only one thing: how much of the task time can AI handle today. You can read how that is built on the Coverage method page.
What has actually been tested
Not much, and that matters. Our evidence grade for quality parity in this occupation is D. A D grade means there is no direct head-to-head test of a machine against a qualified sheet metal worker on this job’s real tasks, so we publish no parity number for it. We would rather say that plainly than put a figure on an untested claim.
What would settle it is specific: a timed trial of a mobile robot fabricating and installing a duct run on an occupied job site, measured on fit tolerance, leak rate, rework and hours, against a journeyman doing the same scope. Bench demonstrations of robotic welding in a fixtured cell do not answer the question, because the fixture is the hard part on site. How we grade evidence is set out in the quality parity method.
The labor market data is steadier ground. The US Bureau of Labor Statistics counts about 119,770 sheet metal workers, with median pay near $61,800 a year and employment projected to grow around 2.5% between 2025 and 2035 (BLS, 2025). That is slow growth, not contraction.
When the picture could change
Most likely after 2048 (8 in 10 of our scenarios). That window reflects how far machine hardware, not just software, would have to come. What the range measures is explained on the replacement-year method page.
Two things could pull it earlier. First, more work moving into the shop: prefabricated duct modules and panelized assemblies built in a controlled cell, then trucked in, shift task time from a chaotic site to a place robots handle well. Second, cheap general-purpose mobile robots. Our cost panel already shows software and tooling for the digital tasks costing a small fraction of the labor they assist, so if capable hardware ever lands in the same range, adoption pressure rises fast.
Two things hold it back. Site conditions are the first: dust, heat, scaffolding, partial power and other trades working in the same ceiling. The second is liability and code. Duct work has to pass inspection, seal to spec and carry the installer’s name; nobody signs an inspection form on a robot’s behalf without a long record. Small contractor payrolls also limit how much capital a shop can put behind an unproven machine.
How to stay needed in this trade
Lean into the tasks the machines cannot reach. Install and troubleshooting work: hanging, sealing and balancing systems in buildings that do not match their drawings. Hand-forming and on-site modification, where a transition gets re-cut to clear an obstruction. And inspection and repair of existing systems, which is diagnostic work nobody can nest into a cut file.
Two skills raise your value either way. One is fluency with the digital layer: model viewers, CAD and CAM, digital measuring tools, and the fabrication software that drives shop equipment. The other is supervision and estimating, because someone has to scope the job, price it and run a crew. Both put you on the side of the tool rather than in front of it.
What to do: ask your shop whether you can get time on the CAM and layout software, since that is the part of the job already changing.
Nearby trades face similar math. Compare this page with Structural Iron and Steel Workers, Boilermakers and Welders, Cutters, Solderers, and Brazers, whose shop work is further along the automation path than their field work. Our Still needs a human score for this job is 83 out of 100 (higher is safer); the full method behind the three questions is at how we score jobs.
From here you can put two trades side by side on the job comparison tool, see where the whole family sits on the construction trades workers page or across construction, and browse the jobs that mostly need a person list to see which occupations share this profile.