Is this the first clear evidence of AI’s impact on the labour market?

Last week we got some striking new evidence that AI may be starting to show up in employment data. On Sunday, US financial analyst Eric Basmajian demonstrated that labour productivity in the sectors with heaviest adoption of AI has decisively ticked up from its pre-2023 trend, while other sectors have not. A day later, a Goldman Sachs research note showed that sectors with higher AI usage are seeing steeper declines in employment. So, are the productivity explosion and “jobpocalypse” upon us?

First up, it’s worth noting a few caveats. The Goldman result is sensitive to the use of a particular data source and looks at a particular time period. Swap in a richer employment dataset and the link with AI disappears. Run the same analysis two years ago and the link with AI comes out stronger, not weaker.

More importantly, if one extends both analyses to a range of other high-income countries and drills down from industries to narrower occupations, the picture becomes decidedly muddier.

First, the link that was visible in the US between AI usage on the one hand, and productivity growth and falling employment on the other, is conspicuously absent elsewhere. Second, where the pattern does show up, it is overwhelmingly a story about the tech sector rather than the broader AI-embracing triumvirate of tech, finance and professional services. Third, the key inflection point comes in 2022 — when AI meant chatbots, not agents. And fourth, the productivity story is not one of accelerating output and revenue, but flatlining or contracting headcount.

We see this at the company level too: whether it’s Meta, Alphabet and Microsoft in the US, or Spotify in Sweden, revenue and headcount were moving in lockstep until around 2022 and then suddenly diverged. Expand the net to finance and professional services, and that pattern is weak to non-existent.

Put all of this together and “productivity is surging in AI-adopting sectors” becomes “rate hikes, inflation and other economic uncertainty starting in 2022 caused companies to pull back on hiring, and unlike other countries and sectors, Big Tech went further in making substantial lay-offs long before production-ready AI”.

It’s a similar story for job displacement. If we look at employment trends in software, the tell-tale signal of entry-level hiring falling away even as senior roles hold firm shows up everywhere from the US, UK and Canada to Germany, the Netherlands and Sweden. But if we look instead at the most AI-exposed jobs outside tech and outside the US, most of those gaps close right up.

To be clear, none of this completely rules out any role for direct AI job displacement. And there is no question that AI is already disrupting the nature of many jobs. It’s also plausible that while hiring slowdowns and lay-offs began as a response to broader economic shocks and uncertainty (some of it AI-related), they could soon or perhaps even now be extended by company beliefs — right or wrong — that AI allows them to do more with less.

But either way the concentration of all the movement in Big Tech and the US should at least tell us that while AI and other economic disruption certainly can shift employment and productivity, the process is heavily contingent and far from automatic. Different countries, sectors, businesses and companies have different structures, norms, employment laws and regulatory environments that lead them to respond differently to the same external force.

AI or not, the US labour market tends to respond more flexibly to economic shocks — its tech sector even more so. And the low-friction and highly scalable nature of software makes Big Tech well positioned to scale up output and revenues with a fixed or even falling headcount in a way that is less true for finance and professional services, where personal relationships still play an important role in expanding output.

Another way of thinking about all of this is that occupational vulnerability to automation is about more than just whether new technologies are good enough at performing particular tasks. The concentration of lay-offs in software tells us businesses where output is highly scalable are especially exposed. As more companies in more sectors essentially become digital services companies, more employers may make the same bet on smaller workforces as US Big Tech.

john.burn-murdoch@ft.com, @jburnmurdoch