Why die work stays on the bench
A tool and die maker turns a drawing into the thing that makes other things: a stamping die, a jig, a mold insert, a fixture that has to hold a part the same way ten thousand times. The planning is the easy half. The hard half is the last few thousandths of an inch. Fitting die halves so they shut clean, stoning a parting line, chasing a burr that only appears on the three-hundredth part. That happens with a file, a stone, a micrometer and a trained hand.
Repair is the other anchor. Dies come back cracked, galled or out of alignment, and nobody hands you a diagnosis with them. The maker reads the witness marks on the part, decides what moved, welds, grinds and re-fits. Each job is a one-off, so there is no large library of repeats for a model to learn from, and no two shops run the same toolroom.
Setup and verification pull the same way. Mounting work in a lathe, mill or surface grinder, dialing it in, then checking dimensions against gauges and prints is a slow loop between hands and eyes. Low volume and high variety are exactly the conditions that make a fixed robot cell expensive and a general-purpose one unproven. That is the core of the answer when people ask whether AI will replace tool and die makers.
What software does, what it assists, and what stays with the toolmaker
The computational slice is already gone, and it went quietly. Our task split puts that share at 0%, which covers work like turning a solid model into toolpaths and simulating a cut before any metal moves, plus the paperwork around tool libraries, job records and maintenance schedules. None of that is the trade itself; it is the desk end of it.
A larger band is assistance rather than handover: 28% of task time. Here software proposes and a person decides. Generative design and simulation can suggest a die layout or predict where a blank will tear. Inspection software can flag a feature drifting out of tolerance across a run. The toolmaker still judges whether the suggestion survives contact with real steel, real springback and a press that has its own habits.
What is left needs a person on the floor: 72% of task time. Hand-fitting and finishing die components, setting up and tending machine tools, and diagnosing a die that has started making bad parts all sit here. So does the quiet work of deciding when a surface is good enough to run and when it is not.
The evidence, and what is missing
Coverage for this job, the share of task time AI can handle today, prints as 12 out of 100; how coverage is measured explains what counts. The quality grade is different. It reads D, and a D grade means no study has yet tested a system against a qualified toolmaker on this job’s own tasks. So we publish no parity number for it, and neither should anyone else.
What would settle it is specific: a benchmarked trial where a machine, with or without an operator, builds and fits a die to print, then repairs a worn one, measured on dimensional accuracy, lead time and scrap against journeyman work. Robotic grinding and deburring papers exist, but they test a step, not the trade. Our scoring method grades evidence rather than guessing at it.
The labor market numbers are firmer. The BLS counted 56,930 tool and die makers in the US with median pay of $64,050 a year (BLS, 2025), and projects employment down 9.2% between 2025 and 2035. That decline is mostly offshoring, consolidation and plant-level productivity, not software doing the fitting.
Good to know: a shrinking headcount and a hard-to-automate task mix can be true at the same time, and here they are.
When this could change
Most likely after 2046 (8 in 10 of our scenarios). The replacement-year method explains how that window is built.
Two things could pull it earlier. First, hardware: most of this job’s exposure is physical, and the robot class implied is a dexterous, general-purpose machine rather than a bolted-down arm. If that hardware gets cheap and reliable, toolroom tasks become reachable. Second, die design and simulation keep improving, which shifts more hours from iteration on the bench to iteration on screen.
Two things hold it back. Setups are one-off, so there is little repetition to amortize a cell against, and the capital cost of a flexible machine still sits far above a toolmaker’s hourly rate for shops of this size. And tolerance judgment is tacit. A maker who knows this press, this steel and this customer carries information no model has been given. You can see both pressures side by side using compare any two jobs.
How to stay needed in a toolroom
Lean into the tasks that sit in the needs-a-person group. Die repair and troubleshooting first: being the person who can look at a bad stamping and name the cause is the most durable skill in the shop. Then precision fitting and finishing, where hand work closes the gap machines leave. Then setup and verification, especially first-article inspection and the decision to run or stop.
Two skills to add. CAM and simulation fluency, so you are the one steering the software rather than receiving its output. And measurement depth: GD&T, CMM programming and gauge design, which makes you the authority on whether a part is right. Shops in manufacturing pay for that authority.
Close trades worth comparing are machinists, CNC tool programmers and patternmakers, metal and plastic. The wider picture sits on the metal and plastic workers family page, and if you want context for where this trade lands against others, see the jobs that most need a person list.