Idle hands 🌴

Almost a hundred years ago, economist John Maynard Keynes predicted 21st century workers would be clocking fewer hours, thanks to productivity gains wrought from technology.

In his 1930 work, Economic Possibilities for our Grandchildren, Keynes said: “Three-hour shifts or a fifteen-hour week may put off the problem for a great while.” Keynes believed we’d still need to work roughly three hours a day to satisfy what he called the “old Adam” of structure and status. But if productivity soars, the “problem” of scarcity fades. What that looks like may become clearer sooner than you think.

Knowledge workers are already grappling with technologies that slash the time it takes to write, read, search, summarise, and code. Will the gains from AI turn into higher GDP and higher pay? Or fewer jobs and a further concentration of capital in the hands of our benevolent and handsome techbro overlords?

SaaSpocalypse now

Agentic AI models have their eye on white-collar work. Earlier this month, enterprise software stocks sold off after Anthropic released industry-specific plug-ins for Claude Cowork. Investors worried agentic assistants could run tasks across whole workflows and undercut the Software-as-a-Service (SaaS) model.

Then, on 23 February 2026, IBM (IBM) dropped sharply after Anthropic highlighted how Claude Code could automate COBOL, a programming language for business. The world of consultants, billable hours and long projects moved with it. Cybersecurity stocks slid, too, after Anthropic announced Claude Code Security, which scans codebases for vulnerabilities and suggests fixes. Learn to code may no longer be sage advice. 

Middling management

AI might do wonders for productivity. In the UK, estimated productivity growth was around 0.58% a year from 2010 to 2024 compared with about 1.9% from 1997 to 2007. The 2008 crash really did a number on us. The UK has a long tail of low-productivity firms and that tail has been getting longer, doubling between 1997 and 2023. Many of these are zombie firms, unproductive companies kept on life support by low interest rates and cheap labour.

Then there’s the management problem. The ONS looked at technology adoption and AI use within management practices. It gave the UK a score of 0.55 (out of 1) in 2023, up from 0.49 in 2020, with a big gap between well-run and poorly-run firms. Only 3% of firms in the bottom decile of management scores said they had tested or adopted AI, versus 37% in the top decile. Bad management is one reason new tools are slow to diffuse throughout the economy.

Despite the grim picture, UK productivity is showing a tentative improvement. The ONS flash estimate for Q3 2025 says output per hour rose 0.7% in the quarter and was 1.1% higher than a year earlier, leaving it 3.1%, above the 2019 average. Whether this is the AI effect is too early to tell. 

USAI

US productivity numbers are punchier than the UK’s. The Bureau of Labor Statistics’ third quarter of 2025 release reported nonfarm business productivity up 4.9%, alongside falling unit labour costs. If output holds up while labour input grows slowly, you get some kind of productivity decoupling. Whether that persists is anyone’s guess but firms are seemingly practicing no hire, no fire.

A survey of 6,000 executives across the US, UK, Germany, and Australia found 70% of firms use AI, yet more than 80% report no impact on either employment or productivity over the past three years. A US study of customer support agents found productivity rose by about 14%, measured by issues resolved per hour. Intriguingly, gains were much larger for novices and lower-skilled workers and smaller for more experienced workers.

Wrung out

New technologies tend to see a lag between adoption and statistics. In the early computer era, for example, productivity gains were slow to appear. This is the J-curve. At first, firms restructure workflows and retrain staff. Only later do the gains scale. We are currently in the spending phase of AI – the dip in the J – where costs are high and the 15-hour-week ROI still over the horizon.

Large language models (LLMs) are very good at the work juniors do: first drafts and first passes. The ONS has put UK unemployment at 5.2% for Oct-Dec 2025, a five-year high. Youth unemployment is higher still, around 16%, the highest since 2014 – including the pandemic – and higher than the EU for the first time. If firms automate away junior work, they may hire fewer people at the start of their careers or demand a higher ready on day one baseline. Ads for entry-level roles, requiring two decades of experience, may become even more common.

Is ghost GDP in the room with us right now?

There’s been a bit of a paranoid sell-off in markets recently, following the publication of a dystopian vision of the near future. The finance fiction suggested AI-driven labour displacement could simultaneously lead to mass white collar unemployment and ballooning corporate profits. This runs counter to most economic models, which show households (especially those of white collar workers) are the primary source of demand in a developed economy.

In a world where firms can produce more with fewer workers, one could see a slower hiring market, weaker wage growth, and a bigger share of national income flowing to profits. Consumption could be propped up by asset-rich households, government transfers, exports, the emergence of different working patterns, or simply lower prices. That could look like solid output while the masses experience insecurity. If AI also drives costs down, it could add another deflationary twist: cheaper goods, but fewer good jobs and less time working. Consider the price of manufactured goods, such as clothing, following the invention of the spinning jenny or the power loom.

Mechanical twerk

Shorter workweeks are not inevitable. The weekend as we know it – Saturday and Sunday off, Mon-Fri labouring – was only popularised in the early 20th century, notably by US labour movements and Henry Ford. You gotta fight for your rights. If gains are captured by capital, saved time does not leisure make.

In the near term, LLMs make chunks of white-collar work faster. They flatten the learning curve for some people. They reduce the cost of producing a decent first draft. We have evidence it can lift productivity in specific settings. Turning those micro gains into macro GDP takes time. Advice for wagies and office drones: expect fewer entry points and higher expectations, and careers reshaped around babysitting bots. At the same time, while AI makes digital goods cheaper, anything requiring the human touch, from surgery to plumbing, gets relatively more expensive.

The devil makes work

If we do get to change the status quo, the risk is always the same mistakes: gains at the top and misery at the bottom. If we automate ourselves out of employment, maybe we can just pretend to work. In China, some unemployed young adults are paying to sit in fake offices and look for jobs. Youth unemployment there sits at around 14%. It’s 16% in the UK, remember. Is the Pretend To Work Company a sign of things to come?

Banishing drudgery is one thing. But work is more than just earning a living. Fewer paths to meaning and purpose have seen work become many people’s raison d'être. How will they cope with near-limitless leisure time? Whichever way you cut it, in the future, it looks like we are going to have a lot more time on our hands. What we do with that time might be one of the defining questions of this century.