Why the work stays on the iron
Ask whether AI will replace steel workers and the honest answer starts with where the job happens: forty feet up, in wind, on a beam that shifted when the crane set it down. Structural iron and steel workers position and secure steel members, bolt or weld the connections, and signal crane operators to land a load within a fraction of an inch. Software can model every piece of that frame. It cannot stand on the steel and feel a connection pull up tight.
The second reason is that no two days match the drawings. Columns come out of plumb. Bolt holes miss by a quarter inch. Someone has to ream, drive a drift pin, or stop and talk to the engineer before the next piece goes up. That judgment call sits between a worker, a foreman and an inspector, and it carries real consequences if it goes wrong. Our scoring treats that kind of accountable, physical decision as work that does not transfer cleanly to a machine. You can read how we weigh it on the methodology page.
The robotics panel above is blunt about the rest. Almost all of this job is physical, and the systems shipping today are fixed automation: bolted down in a shop or plant, fed identical parts, fenced off from people. An open structural frame is the opposite of that. The ground changes, the weather changes, and the next lift is never quite the last one.
What software takes, what it assists, what it leaves
Planning and record work is where tools already take over. Member lists, quantities and shop tickets come straight out of a 3D model. Sequencing and delivery schedules get built and rebuilt without anyone redrawing them. That slice of the day, about 8% of task time, is the paperwork end of the trade. The coverage score explains how that share is measured.
Assistance shows up right next to the hands. A tablet model replaces rolled prints when crews verify that uprights are plumb and level. Scanners compare the erected frame against the design before the deck goes on. Machine vision checks shop welds, and sensors flag fall and load hazards. Roughly 0% of task time is assisted work, where the tool speeds up a person who is still doing the job.
The remaining share, 92%, stays with people. That is the trade itself: connecting structural members and bolting or welding them in place, cutting and fitting steel to field conditions, and rigging loads and signaling the crane. Add erecting metal frames and tanks on site, and you have most of a shift that no current system performs unsupervised.
What has actually been tested
Not much, and that matters. Our evidence grade here is D on an A to D scale, and D means there is no direct, published test of an AI or robotic system against a qualified ironworker in our evidence set. So we publish no parity number for this job. A grade without a measurement is more honest than a figure with nothing behind it; the quality parity method sets out the rule.
What would settle it is specific. A field trial where a machine erects and connects structural steel on a live job, with reported first-time fit rates, rework hours, inspection pass rates and injury data. Or a shop benchmark where a robotic cell fabricates and fits assemblies against a certified fitter, measured on tolerance and scrap. Until something like that is published and repeated, the honest position is an untested one.
For context on the market rather than the machines: about 68,380 people hold this job in the US, median pay is $62,780, and employment is projected to grow 3.1% between 2025 and 2035 (BLS, 2025). That is steady demand, not a shrinking trade.
When this could shift
Most likely after 2048 (8 in 10 of our scenarios). The replacement-year method explains exactly what that window covers and how the range is built.
Two things could pull it earlier. First, prefabrication: every connection made in a controlled shop is a connection a robotic welding cell can reach, and it moves hours off the site. Second, cheap sensing. Drones, scanners and tablet models are already cutting layout and inspection time, and each gain there chips at the non-physical part of the day.
Two things hold it back. The automation tier is the big one: fixed systems need a repeatable setting, and steel erection offers the opposite. The other is sign-off. Structural connections are inspected, certified and insured, and somebody qualified has to put a name to them. The cost panel above shows how far apart machine and human hours sit for this work today, and site-grade equipment does not get cheaper quickly.
What to do: treat the shop side of the trade as the part most likely to change, and keep your field and rigging credentials current.
How to stay needed in this trade
Lean into the work the task list leaves with people. Connecting and rigging is the first: landing, aligning and securing members under a crane is the skill the whole job turns on. Field fitting is the second: cutting, reaming and welding steel to conditions nobody drew. Third is reconciling the model against what is actually built, then deciding what to do about the gap.
Two skills compound. Keep a structural welding qualification current, since certified field welds are hard to hand to anyone else. Add rigging and crane signaling, which puts you in charge of the lift rather than beside it. Reading 3D models fluently is now part of both.
If you want to see how close work scores, look at reinforcing iron and rebar workers, structural metal fabricators and fitters, and welders, cutters, solderers and brazers. The fabrication and shop roles sit closer to fixed automation than field erection does. You can also browse the construction trades family, read the wider construction sector picture, put two trades side by side on the compare page, or see where this job lands among the jobs that mostly need a person.