X discussion: FT chief data reporter John Burn-Murdoch says the US productivity jump "doesn't look like AI" and is mostly tech layoffs, while Gavin Baker argues it shows "ROI on AI"

Background

Whether AI is actually raising productivity is one of the most argued questions in investing and policy this year. A chart showing US tech, finance and professional services pulling far ahead of the UK, Canada, Australia and Europe in output per worker has been circulating on X as proof that AI is showing up in the macro data. FT chief data reporter John Burn-Murdoch says he made that chart, and that he made it to argue the opposite.

The argument

His reasoning has three steps. First, timing: the rise starts in 2020, and even a 2022 re-acceleration predates usable AI agents. Second, geography: these sectors use AI just as widely in other rich countries, which show no similar rise. Third, decomposition: output grew roughly on trend, and the change comes mostly from US tech layoffs and hiring pullbacks. He calls it the old story of the US labour market reacting more flexibly to the same shocks. Gavin Baker argues instead that the reversal of pandemic over-hiring began before ChatGPT, and the upward trend resumed afterward, which he reads as a sign of returns on AI.

Benchmarks and data

There are no model benchmarks here. The numbers come from the two posts: Baker cites EPB Research saying the 3 most AI-heavy industries went from 2.2% to 5% productivity growth, and says the US is 72% of Anthropic revenue. These are third-party or vendor figures he cites, and AICoder has not verified them. On reach, Burn-Murdoch's opener has about 186K views and Baker's reply about 102K.

What it means for builders

For teams selling AI products, the takeaway is that macro data cannot yet prove AI's productivity effect on its own. A more reliable way to show value is to measure it in your own setting: task completion time, rework rate and output per person, before and after. For the full argument, read Burn-Murdoch's thread and the FT piece he linked (paywalled).