How not to lose your job to AI

Benjamin Todd argues automation doesn’t uniformly push wages down. Partial automation often raises the wages of whatever skills are left, because those skills become the new bottleneck: when ATMs arrived, bank tellers per branch dropped from 21 to 13, but cheaper branches meant banks opened more of them, and total teller employment grew for two decades, with tellers spending their time talking to customers instead of counting money. It was only once online banking automated the job further that employment actually fell. The question isn’t which job title survives AI, but which skills become scarce as AI absorbs the rest.

From that pattern he builds a framework for which skills gain value, then applies it to name six worth building now, while flagging coding, routine white-collar work, visual creation, and predictable manual labor as the categories with the most uncertain outlook.

Fichamento

#

• Partial automation has repeatedly raised total employment even while cutting the wage of the automated task itself: British textile manufacturing grew employment during the industrial revolution despite heavy automation, and global translator employment is still up net of AI’s effect on the profession.
• Full automation is a different regime. Epoch AI’s GATE model shows wages rising roughly tenfold as non-automated tasks become bottlenecks, then crashing once the final bottlenecks are removed. If humans keep even 1% of tasks, the same model shows wages rising indefinitely instead.
• Four traits predict which skills gain value as AI advances: hard for AI to learn (messy, long-horizon, no clean training data), needed to deploy AI itself (auditing, directing, building the infrastructure around it), tied to goods people would consume a lot more of if they got cheaper, and hard for other humans to pick up quickly.
• People-in-the-loop persists for reasons beyond raw AI capability: legal liability, demand for high reliability, professional lobbying, preference for a human touch, required physical presence, institutional inertia, and the need to specify what humans actually want.
• The six skill clusters Todd names as good bets: using AI to solve real problems, personal effectiveness (productivity, social skills, learning how to learn), leadership (entrepreneurship, management, strategy, true expertise), communications and taste, getting things done in government, and complex physical skills in unpredictable, high-demand environments such as data centre construction.
• Skills flagged as most uncertain: routine white-collar work (writing, admin, analysis, translation), coding and applied STEM, visual creation such as animation, and predictable manual labor such as driving. He expects white-collar jobs around the 70th-90th income percentile to be hit hardest, with organizations becoming more top-heavy as fewer people oversee more AI agents.
• Career advice for someone early in a white-collar track: look for ways to skip routine entry-level roles (smaller, faster-growing organizations, side projects, managing a contractor on a small scale), be cautious about multi-year training like PhDs or medicine, and build general resilience through savings, mental health, and not tying yourself to a single place.