Ai is eating white-collar jobs but can't fix your sink
Javier García, 34, left his data-analyst desk in Madrid last November, bought a €120 set of pliers and a blowtorch, and now bills €90 an hour unclogging pipes. His former co-workers, the ones still debugging Python scripts for half that wage, call him crazy. García calls them redundant.
The twist is brutal: the smarter the algorithm, the safer the wrench. While generative models churn out annual reports, legal briefs and even lines of code in milliseconds, they still struggle to distinguish a Phillips from a flat-head screwdriver. The result is a labour market flipping faster than a blockchain valuation.
The numbers show who survives
Harvard Business School tracked 1,000 medium-skill occupations across 12 countries. Jobs that rely on structured data—book-keepers, payroll clerks, entry-level paralegals—shed 18 % of positions in two years. Meanwhile, demand for elevator repairers jumped 11 %, wind-turbine technicians 14 %, specialised welders 9 %. The pattern is mechanical, not metaphorical.
Electrician James Vandall summarises it with a grin: 'AI can write you a poem about electricity, but it can't strip a live wire without killing itself.' His union in Ohio reports apprenticeship applications up 37 % since ChatGPT went mainstream. College enrolment in electromechanical engineering is rising for the first time in a decade; computer-science departments are quietly trimming intake forecasts.
Why the resilience? Physical trades operate in what roboticists label 'unstructured environments'. A plumber arrives at a 1950s flat to find a copper pipe retrofitted with three generations of mystery joints, a nest of towels, a furious cat and a landlord who insists nothing is broken. No data set captures that chaos. Training a neural net on every possible corrosion pattern would cost more than hiring every human plumber alive.
The hidden premium on thumbs
Researchers at MIT estimate the cost of a general-purpose repair robot at roughly $2.3 million per unit, assuming a five-year R&D horizon. A seasoned HVAC contractor, tools included, costs $0. Add the liability insurance for a 30-kilo steel arm swinging a blowtorch in a wooden attic and the economics collapse. Venture capital has noticed: funding for 'physical AI' startups dropped 28 % last quarter while hand-tool makers like Milwaukee and DeWalt post record sales.
Even the flagship factories tell the same story. Tesla's Fremont plant still employs 600 welders per shift because the Cybertruck's aluminium curves exceed the precision tolerance of current robots. BMW's Spartanburg facility flies in German sheet-metal masters to teach cobots how to 'feel' micro-variations in alloy tension. The humans aren't being replaced; they're being flown business class.
White-collar workers are scrambling for a Plan B. LinkedIn Learning lists 'basic carpentry' as the fastest-growing elective among software engineers. Amazon's internal career portal shows 4,200 coders enrolled in night courses for electrician certification, up from 14 two years ago. The joke inside Seattle headquarters is that the next Alexa will be installed by the same people who built her.
The overwork paradox nobody predicted
Yet the migration has a cost. Emergency-room visits for DIY accidents rose 22 % in Germany last year; the health ministry blames YouTube tutorials and overconfidence. Meanwhile, master tradespeople raise hourly rates faster than central banks hike interest. In Paris, a licensed plumber charges €180 after hours, more than a junior lawyer. London bricklayers report six-month waiting lists; one crew recently accepted payment in Bitcoin to jump the queue.
Policy makers are waking up late. The EU will spend €3 billion retraining 200,000 displaced office workers as solar-panel cleaners and heat-pump installers. Japan subsidises vocational schools that promise to enrol humanities graduates within 90 days of layoff. The irony stings: societies that spent 20 years pushing every teenager toward university now import Filipino welders on special visas.
For García, Sundays are the new bonus day. He finishes a boiler retrofit in a six-floor walk-up, pockets €400 cash and scrolls through former colleagues' LinkedIn posts about 'AI-first synergies'. He no longer comments. 'They'll figure it out,' he says, wiping solder flux from his hands. 'Right after the algorithm learns to crawl under a sink.'