More AI, less free time?

15 October 2026

Many people hope that AI will take work off our hands. The four-day week seems within reach. An Instagram post by @dopameme.co argues against this: the AI productivity paradox predicts rather the opposite.

The story of farming

The post recalls the invention of agriculture. Back then, so the story goes, people thought: sow once, eat forever, finally rest. Instead, hunter-gatherers supposedly worked fewer hours and were healthier than the first farmers.

Whether that is true in detail is disputed among experts. The core is familiar all the same: when work gets faster, people often do not work less. More is expected.

Who pays for all this?

A second post raises the money question. It draws on a study by Stijn Van Nieuwerburgh (Columbia Business School), published by the Brookings Institution. It estimates investment in AI infrastructure in the USA from 2025 to 2032 at around 10.3 trillion dollars. That is on average about 3.6 per cent of economic output per year. A version of the same paper at the NBER arrives at just under 9 trillion.

This means data centres, power supply, networks and chips. Relative to the economy, it would be more than earlier big build-out waves such as railways, electrification or motorways. The study warns: a growing share is paid for through loans and nested financing. This makes risks harder to see.

For this to pay off, AI has to earn a great deal of money. And in the end, that money comes from us: from companies, customers, jobs.

Two possible endings

How far can this go? The scenario AI 2027 describes a future in which AI becomes superhumanly capable within a few years. It was written by Daniel Kokotajlo and colleagues and published in April 2025. There are two endings: one with a slowdown, one with a race. The authors themselves stress that 2027 is not a fixed date.

You do not have to believe the scenario. But it helps you to check your own assumptions.

What I take away from it

AI makes me faster. Whether that gives me more time is not up to the AI. That is up to me, or to someone else deciding for me.

A nice thought on this came with another post: a life has around 4,700 weeks. The question is not how much of it AI saves. The question is what I use the saved time for.

Sources: Instagram @dopameme.co · Instagram @askesium · Brookings: Financing the AI Buildout · NBER Working Paper 35865 · AI 2027 · Instagram @schraffuren