General Purpose Technologies

Most technologies do one job. A pacemaker steadies a heart; a barcode scanner reads barcodes. Economists reserve the term general purpose technology for a rarer kind: one so broadly useful that it eventually reshapes the entire economy. The canonical examples are the steam engine, electricity, and the computer. Three traits set them apart:

1. they can be applied almost anywhere,
2. they keep improving for decades, and
3. they enable further inventions nobody planned for.

Electricity wasn't just a better way to light a room — it made possible the assembly line, refrigeration, and radio.

AI is widely considered the newest member of this small club. If that's right, history offers two lessons worth taking seriously.

First, the payoff arrives late and unevenly. American factories began electrifying in the 1890s, but measurable productivity gains took roughly thirty years to appear. The reason: early adopters simply swapped a big electric motor in for the old steam engine and kept everything else the same. The gains came only when factories were rebuilt around what electricity made newly possible — small motors at each workstation, single-story layouts, work organized around the flow of materials instead of the drive shaft. The technology was necessary, but the reorganization was where the value lived.

Second, that lag is where the risks concentrate. For a business, the most common failure isn't ignoring the technology, but rather investing in it without changing how work gets done. This produces impressive pilots and no results. The opposite failure is real too: waiting for certainty while a competitor rebuilds their operations and pulls ahead. And because general purpose technologies touch everything, they are genuinely destabilizing at larger scales — displacing jobs and skills, shifting power toward those who control the technology, and creating new failure modes faster than institutions adapt. Prudence and ambition both have their place.

The practical question, then, is rarely "should we adopt AI?" It is more specific and harder: which of your workflows can this technology transform first, what would you rebuild around it, and how will you know whether it's working? These are all questions we can answer together.