EXPOSED
Disassembling a mechanical movement, cleaning and lubricating jewels, adjusting hairsprings, and fabricating or fitting replacement parts for a 1920s clock are micro-manual tasks no robot performs at commercial cost, and the diagnosis happens through touch, sound, and timing-machine traces rather than text. The genuine AI exposure is thin and peripheral — parts lookup, movement identification, repair-cost estimating, and customer correspondence. What holds this occupation back is not AI but market size: 1,310 jobs, quartz and smartwatch replacement culture, and demand concentrated in luxury service and heirloom restoration where the named craftsperson is part of what the customer buys.
Tasks largely resist digitisation. Poising a balance wheel on a staking set, bending a hairspring back to flat with tweezers under a loupe, and turning a replacement pivot on a Jacot tool are tasks where the tolerance is microns and the workpiece is unique to that caliber — 17 rather than 20 only because catalog identification, movement dating, and quoting are drifting to software.
Hands-on in uncontrolled environments. The work is entirely in the hands: pegging out jewel holes, listening to a beat rate, feeling end-shake in a train bridge, and every bench is a one-off environment of non-standard vintage parts rather than a fixtured assembly line — held below 19 because it's a stationary indoor bench, not a job site, so the physical variability is in the object, not the surroundings.
No licence, no signature requirement. No state licenses watchmakers; WOSTEP, SAWTA or AWCI CW21 certification opens doors at Rolex or Swatch service centers but nothing in law stops an uncertified person from opening a caseback, and the shop's liability for a damaged heirloom runs through business insurance, not a personal license.
The human relationship is the product. Customers hand over a grandfather's Hamilton or a $30k Patek to a specific named bench worker they were referred to, and repeat service intervals every 5-7 years make it a decades-long relationship — short of 17+ because factory-authorized centers and mail-in depots do capture volume anonymously under a brand name.
Meaningful discretion. Deciding whether to replace an original mainspring or preserve it, whether refinishing a dial destroys collector value, and when a worn barrel arbor is repairable versus scrap are irreversible calls made without the owner watching — but they run on established restoration convention and a quoted estimate, not open-ended high-stakes ambiguity.
Has AI actually changed your work?