AI and work

From steam to computers

Take steam. In the early 18th century, steam engines were clever but narrow tools, mainly used to pump water out of coal mines. They did that far better than horses and buckets, but the wider economy barely noticed. Factories still ran on water wheels. Households felt no change. Had steam stayed in the mines, it would have been a mere footnote in the history books, not a transformative force. Productivity only surged when steam engines were put on wheels and rails. Railroads slashed travel times and freight costs, rewrote trade routes and labour markets, and even forced the standardisation of time itself. Steam’s economic impact changed when its application spread—from a stationary pump underground to a moving, networked transport system above it.

Electricity followed a similar pattern. At the turn of the 20th century, it was mostly treated as a replacement: swap steam engines for electric motors, gas lamps for light bulbs, and carry on. That helped—motors were more flexible, bulbs safer and cleaner—but for a while electricity looked like a useful, incremental upgrade, not a revolution. The real transformation came when electricity found new applications that reorganised production, cities, and homes. Factories rebuilt around small, decentralised motors instead of a single great engine, enabling new layouts and higher output. Streetcars and subways turned electricity into urban mobility, stretching commuting distances and reshaping cities. In households, reliable power brought fridges, washing machines, vacuum cleaners, irons, radios, and televisions. Each was just another application of the same technology; together they changed daily life and opened vast new markets.

The internal combustion engine is another case in point. On its own, it is just a machine that turns fuel into motion—useful, but not transformative. Its real economic power appears when you follow it into its applications. First, it powered cars. The automobile did far more than replace the horse and carriage: it reshaped where people lived and worked, fuelled suburban growth, and spawned roadside retail and new commuting patterns. The same engines went into trucks, enabling long-haul freight, and just-in-time manufacturing. Then the engine took to the skies. Air travel turned rare journeys into routine trips and expanded tourism on a scale 19th-century observers could hardly imagine.

Computers and the Internet repeated the story at higher speed. For a long stretch, computers sat in back offices, running payroll systems and inventory management. They were valuable tools, yes, but mostly in the business of streamlining existing processes. The dramatic productivity gains of the late 1990s and early 2000s came when computing was applied in new ways: networked PCs in offices, of course, but also e-commerce websites, digital marketplaces, search engines, and later smartphones and app stores.

Where does that leave AI?

At the moment, much of today’s AI is being used in a familiar way: as a powerful assistant inside the existing workflow. It helps workers draft emails, polish marketing copy, write code, summarise long documents, and answer customer queries. These are worthwhile applications. Early evidence shows large time savings, especially on routine, text-heavy tasks. Workers report that they can get through their inbox faster. Junior staff can produce higher-quality first drafts. Support agents can handle more queries per hour.

If AI were to stop there, it would still matter. Just as early electrification or early computing had clear benefits, so too does this wave of AI as a productivity tool. But history suggests that the real economic payoff comes when a technology is applied in places that have long resisted productivity improvements.

Think about the service sector. For decades, economists have worried about what William Baumol famously called the “cost disease” of services. Many services—care work, cleaning, hospitality, basic retail, everyday logistics—have been hard to automate and slow to grow in productivity. A teacher still stands in front of a classroom. A waiter still carries plates to the table. A care worker still helps someone in and out of bed. These jobs are essential but labour-intensive. As a result, a large share of the modern economy has remained stuck with low productivity growth.

AI has the potential to change this, not just because of its capacity to automate knowledge work, but because of very concrete applications that pair AI with sensors, actuators, and mobile hardware. In other words: robots.

For decades, robots have lived mainly in factories, welding car bodies and assembling electronics on carefully choreographed production lines. These environments are controlled and predictable, which suits traditional industrial robots well. But they represent only a sliver of the overall economy. What is now becoming possible, thanks to recent advances in AI, is a different type of robot: one able to navigate cluttered spaces, recognise objects, interpret instructions in natural language, adapt to changing conditions, and coordinate with people.

As these robots make their way into supermarkets, warehouses, hospitals, hotels, and homes, we should expect the productivity story of AI to shift up a gear. A cleaning robot that can operate safely alongside people in a shopping centre or airport is a new application of AI. So is a robot that can restock shelves, move goods through a warehouse without fixed tracks, or assist a care worker by handling the more physically demanding parts of the job. Add in autonomous or semi-autonomous vehicles, drones for deliveries and inspections, and AI systems coordinating fleets of such machines, and you begin to see how the technology’s reach could extend deep into the low-productivity heartland of the service economy.