🔮 Task first, pay later: AI’s twin routes for wages and work

Routine automation trims headcount; complex automation trims salaries

An earlier version of this post incorrectly stated that Recruit Holdings cut 1,300 HR and recruiting roles. In fact, the layoffs affected roles across a range of departments and functions. The edition below has been updated to reflect this more accurately. Thanks to EV member Tom W. for flagging the error.

---

A Mad Max economy, where your hard-earned expertise trades at commodity prices, is a plausible future of work. One type of AI automation deployment could move millions into low-paid service roles even as employment survives on paper.

However, that same technology could also have the opposite effect. It could turn scarce, high-paying jobs into mass-market opportunities. Furthermore, create jobs that don’t yet exist.

MIT economists David Autor and Neil Thompson explore what is at stake in their landmark study. After analyzing four decades of data across 303 US occupations, they found that AI’s wage effects hinge on one crucial factor. That factor is what gets automated. Not whether firms adopt AI, but which tasks they hand over to machines.

Let’s look into this further.

A Mad Max economy is a choice, not our fate. It’s a good movie, though.

Two pathways 

When firms automate the complex bits of a role—say, the knowledge or judgment-heavy tasks—wages tend to fall, while employment increases. Autor and Thompson's data shows that a one-standard-deviation drop in task expertise over a decade is linked to an approximate 18% wage decline while employment roughly doubles.

Telephone operators from 1980-2018 are the canonical case, where technical simplification made it easier for more people to enter, but at a lower pay.

And Uber is a modern parallel. By breaking the stranglehold that medallion owners once held over New York’s taxi market, the mix of GPS routing and app-based matching opened ride-hailing to a far broader pool of drivers and passengers. Between 2014 and 2024 the city’s total ride market, measured by fares, expanded 228%, while the number of active drivers nearly doubled from 44,000 to 95,000. This greater supply cut average earnings – after allowing for roughly 32% inflation, a typical yellow-cab driver earned 10-15% less in real terms. Notably, about two-thirds of former taxi drivers left the sector altogether, with some transitioning to drive for Uber and others pursuing alternative employment.

The other path looks very different.

Several major firms are now embracing routine-task automation, targeting predictable and codified tasks like call handling, screening, scheduling. The outcome flips: employment shrinks, but wages rise for those who remain. BT plans to cut up to 55,000 jobs – about 40% of its workforce – by 2030, as fibre roll-out and generative AI take over routine customer service work. Recruit Holdings, which owns Indeed and Glassdoor, just cut 1,300 roles after embedding large-language models across its platforms. At Amazon, where robots already outnumber human staff, CEO Andy Jassy admits that generative “agents” will eliminate some job families even as they create new technical ones. In each case, automation strips out low-complexity work. Headcount contracts, the skills bar rises and the wage ladder steepens. This may already be the case for AI coding roles, with sky-high salaries for super-coders, widening the gap between them and the average coder.

Yet this outcome isn’t inevitable.