Why one-off parts keep this work with people
A model maker builds the first physical version of a part. Not a hundred of them. One. That one part has to be cut, fitted and proved before anyone commits to tooling. Software can draw it and plan the cuts, but someone still has to set up the lathe or mill, hold the work, watch how the metal or plastic behaves, and correct the plan when it doesn’t match the drawing.
Two tasks carry most of the weight. The first is setup: choosing stock, fixtures and cutters, then dialing the machine in so the first cut lands where the print says. The second is fitting and finishing by hand, filing, scraping, sanding and checking with micrometers and gauges until the piece assembles with the rest of the model. Both depend on touch, sound and judgment built over years.
The job is also small and shrinking for reasons that predate current AI. About 2,610 people hold this title in the US, with median pay of $63,340 (BLS, 2025), and BLS projects employment falling 17.3% between 2025 and 2035. Consolidated prototype shops, outsourced work and 3D printing have been reshaping it for a long while. Fewer openings is a different problem from the craft disappearing, and it’s the one most model makers will meet first.
What software does, what it assists, and what stays in the shop
Work software can take end to end is mostly paperwork and planning around the bench: generating toolpaths from a CAD file, checking a model for obvious geometry errors, estimating stock and cycle time, and writing up job notes. The task split above puts the machine-only share at 0%.
Assistance is where most of the change shows up. CAM software now suggests cutting strategies and speeds a programmer would once have worked out alone, and vision-based inspection can flag a dimension that drifted before a person reaches for the gauge. Our assisted share sits at 23%. The model maker still signs off; the tool shortens the route.
What stays with people is the physical and the diagnostic. Holding an awkward part without distorting it. Deciding that a prototype’s wall is too thin to machine as drawn and calling the designer. Hand-finishing a surface until it feels right. That people-only block comes to 77%, and you can see how our coverage score is built from those task shares.
What has actually been tested
Not much, in this job specifically. Our evidence grade here is D, which means no study has yet put an AI system against a working model maker on this job’s own tasks and measured the result. So we publish no parity number for it. Claiming one would be guesswork dressed up as data.
Two kinds of test would settle it. One: a timed trial where a system takes a drawing, programs the job, runs the machine and delivers an inspected part within tolerance, scored against an experienced maker doing the same work. Two: a measured comparison of AI-assisted CAM programming versus a skilled programmer on setup time, scrap rate and first-part accuracy. Until something like that exists, the honest answer is that the hands-on side is untested, not proven safe. Our full scoring method explains how grades move when evidence arrives.
When the picture could shift
Most likely after 2046 (8 in 10 of our scenarios). We explain what that window measures on the replacement-year method page.
Two things could pull it earlier. Metal and high-grade polymer printing keeps improving, and every prototype that comes off a printer instead of a mill is machining time that never gets booked. Robot cells that handle their own loading, fixturing and tool changes would take over the repeatable half of a setup, leaving the maker to the tricky parts.
Two things push it back. The automation that fits this shop floor is fixed automation: it pays for itself over a production run, and a one-off prototype has no run. And the workforce is tiny, so there is little commercial reason to build a dedicated system for it when the same robotics money serves a line running thousands of parts. Hand fitting and judgment calls on a part nobody has made before remain stubborn problems for machines.
Good to know: the risk to this job over the next decade looks more like fewer shops and fewer trainee benches than like a robot doing the whole build.
How to stay needed
Lean into the three tasks that machines handle worst here. First, setup and workholding on unfamiliar parts, where the fixture is half the problem. Second, hand fitting and finishing to a tolerance that no program specified. Third, design-for-manufacture conversations: telling an engineer early that the part as drawn cannot be cut, and offering the version that can.
Two skills compound. Multi-axis CAM programming, so you own the software step rather than waiting on it. And metrology: CMM work, GD&T and inspection reporting, which is the skill that certifies everyone else’s output, including a machine’s.
If you are weighing a move, the closest work sits nearby. Compare this job with Patternmakers, Metal and Plastic, Tool and Die Makers and Model Makers, Wood, or look across the whole metal and plastic workers family and the wider manufacturing sector. You can also put two jobs side by side on the compare tool, or see where hands-on trades land in our list of jobs that mostly need a person.