Industries / Mining and materials / Steel and metals / Tool and die
Passing on the skill
Passing on the skill is how toolmaking is taught, tested and handed from experienced toolmakers to new ones: apprenticeships, national skill tests, contests and, lately, software tutors.
5
Milestones achieved, from 1950 to 2018.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1950
Skills judged in contest
Francisco Albert Vidal organized a skills contest in Madrid in 1950 for young people from Spain and Portugal. By 1953 competitors came from Germany, Great Britain, France, Morocco and Switzerland, and the contest grew into WorldSkills. CNC milling, cutting metal on a computer-controlled milling machine, is one of the skills it tests, and WorldSkills Shanghai 2026 awarded medals in it.
Credit
Francisco Albert Vidal; WorldSkills International
1959
Japan tests die finishing
Japan's national trade skill test (ginō kentei) rates how well a worker has mastered a trade, and those who pass may call themselves ginōshi, certified skilled workers. The labor ministry's chart of tested trades starts in 1959 and lists finishing of dies and of jigs and tools from that year. Die making itself, for press dies and for plastic molds, is a separate tested trade today. As of March 2026 the test covered 133 trades.
Credit
Japan's Ministry of Health, Labour and Welfare
1987
Germany redefines the toolmaker
On 15 January 1987 Germany issued new training rules for the industrial metal trades, merging many narrow trades into a few broad ones. The old trades, among them the Werkzeugmacher (toolmaker), one of the most common in 1986, were phased out with transition periods to 1991. The new Werkzeugmechaniker (tool mechanic) trained in one of three fields: stamping and forming, molds, or instruments. The employers' federation Gesamtmetall and the IG Metall union had agreed the principles in 1977 and 1978, aiming at a highly skilled worker able to act independently.
Credit
Gesamtmetall and IG Metall; Federal Institute for Vocational Education and Training (BIBB)
1995
Skills judged on a part
The National Institute for Metalworking Skills (NIMS) was formed in the United States in 1995 and writes skill standards and credentials with industry groups, including the National Tooling & Machining Association and the Precision Metalforming Association. Most of its credentials pair a written exam with a performance evaluation. In one type, the candidate machines a part designed by NIMS to its drawing, and the finished part is inspected.
Credit
NIMS
2018
Digital work enters training
From 1 August 2018 eleven German industrial metal and electrical trades, the tool mechanic among them, must teach "digitalization of work, data protection and information security" throughout training. The metal trades also gained optional add-on qualifications, among them additive manufacturing, each about eight weeks long and examined in a case-based discussion. Today the tool mechanic's training takes 42 months and covers four fields: molds, instruments, stamping and fixtures.
Credit
BIBB, with employers and unions, for the federal government
Watch for
AI checks student G-code
Adam Hathaway, who teaches machine tool technology at Chabot College in California, built AI assistants on the Nectir platform that check students' G-code, the instructions a CNC machine follows, line by line against his own standards and explain each error. He says his students' programs and test results improved, and that the setup took about 20 hours over a year.
Achieved whenthe college or an outside evaluator publishes a comparison of course sections with and without the assistant, with error counts and test scores.
Claimed by
Nectir, quoting Hathaway, February 2026
Five AI agents per machine
Singapore's Institute of Technical Education (ITE) says its AMCAM system teaches CNC machining with five AI agents working beside students at real machines: one builds a digital twin from calibration cuts, one predicts how a chosen feed rate or depth of cut will behave, one tunes settings, one watches the machine's sensors and one guides the learner's next step. The project won a Learning Technologies Gold Award; no learning results were published.
Achieved whenITE or an outside evaluator publishes learning results against students taught without it.
Claimed by
ITE, January 2026
Open
01
A casebook of worn tools
An open, photographed casebook of real worn and broken tools from die and mold shops, such as chipped punches, galled draw dies, worn trim steels and eroded mold gates. Each case carries a toolmaker's written diagnosis of the cause and the fix, checked by a second toolmaker. Trainees use it to learn to read damage, and it doubles as a test of AI vision-language models.
The briefCredit · unclaimed02
Ask the retiring toolmaker
A filmed, searchable record of how experienced toolmakers do named tasks, such as spotting a die, setting punch clearance, stoning a draw radius or finding why an ejector sticks, with their spoken reasoning. An AI model transcribes and indexes the films and answers a trainee's question by pointing to the right clip and quoting the toolmaker.
The briefCredit · unclaimed
Open 01 · the brief
A casebook of worn tools
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, photographed casebook of real worn and broken tools from die and mold shops, such as chipped punches, galled draw dies, worn trim steels and eroded mold gates. Each case carries a toolmaker's written diagnosis of the cause and the fix, checked by a second toolmaker. Trainees use it to learn to read damage, and it doubles as a test of AI vision-language models.
Why it is still open
The labeled tool-damage images found come from lab presses: TU Darmstadt captured 125,000 images of blanking punches and labeled 309, and no public release was found. Engineering question sets for AI, such as MechReason (12,000 questions drawn from research papers, 2026) and ManuRAG's 1,515 manufacturing questions, are not built from shop-floor tool damage.
Built with what exists
A phone with a clip-on macro lens or a USB microscope; Label Studio or CVAT for marking the damage; current vision-language models through their APIs for the test; a scoring sheet filled in by three toolmakers.
First step
Ask one die shop to let you photograph 30 tools in its repair queue, and have its lead toolmaker write a two-line cause and fix for each.
Done looks like
500 cases from at least 5 shops covering at least 8 kinds of damage, under an open license; the agreement rate between the two toolmakers' diagnoses; and a table giving at least 3 AI models' share of correct causes and fixes, as scored by the toolmakers.
Open 02 · the brief
Ask the retiring toolmaker
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A filmed, searchable record of how experienced toolmakers do named tasks, such as spotting a die, setting punch clearance, stoning a draw radius or finding why an ejector sticks, with their spoken reasoning. An AI model transcribes and indexes the films and answers a trainee's question by pointing to the right clip and quoting the toolmaker.
Why it is still open
Oral history archives record toolmakers' careers rather than their methods; a 2005 Michigan State University interview with a Fisher Body toolmaker covers his hiring, apprenticeship, safety and union work. The CNC teaching assistant at Chabot College is built from an instructor's written standards. In Japan, LIGHTz turns veterans' reasoning into diagrams and written procedures as a consulting service; its September 2025 talk shows written templates, not filmed hand work. Kiko Mold's 2026 project in Gifu has AI draft finishing steps from written correction sheets and past records, not from filmed hand work.
Built with what exists
Two cameras, one head-mounted; a lapel microphone; a speech-to-text model such as Whisper or ElevenLabs Scribe v3 (see the Note "Fewer word errors"); a long-context AI model such as Claude or Gemini to index the transcripts and answer with timestamps; toolmakers to grade the answers.
First step
Film one toolmaker sharpening a punch or spotting a die for an hour while narrating, transcribe it, and ask three apprentices to write ten questions about it.
Done looks like
At least 20 hours of film covering 30 named tasks by at least 5 toolmakers, published with their consent, and 100 real trainee questions answered by the system, with the share that toolmakers grade both correct and correctly cited.