Despite decades of technological revolutions, American economic growth has remained remarkably stable at around two per cent per year on average for 150 years.
Stanford economist Chad Jones has highlighted a striking paradox: computers today have 100 million times more transistors than in the 1970s, yet researchers may be only two or three times more productive.
This puzzle sits at the heart of how economists and policymakers are trying to understand what AI will actually mean for the broader economy over the coming years.
Lewis Liu, co-founder and CEO of Twin1 AI, argues the answer lies in what Jones calls the “weak links” theory, which has significant implications for how businesses are structured.
Jones uses the iPhone as an illustration: the entire production chain, spanning design, sourcing, manufacturing, logistics, shipping, retail and marketing, can be undermined if any single part fails.
Making one part of that chain dramatically more efficient helps, but the whole system remains constrained by its slowest or weakest components, no matter how advanced other parts become.
Liu offers a telling real-world example, describing a client who purchased an HR workflow AI agent designed to reduce headcount by roughly ten per cent by automating one of ten tasks an HR officer performs.
The result was that the HR team felt threatened, the company spent a significant amount of money on the AI agent, costs actually went up, and the team stayed exactly the same size with no productivity gained.
Historical precedent suggests this pattern is familiar, pointing to the transition from steam to electricity in manufacturing as one of the most instructive technological shifts in economic history.
Early factories often simply replaced steam engines with electric motors while leaving the existing paradigm intact, capturing only a fraction of what electricity could actually deliver in productivity terms.
The real transformation came only when electricity diffused throughout factories and smaller motors began powering individual machines, allowing factory floors to be completely reorganised around workflow rather than around the limitations of a central power source.
Jones argues that AI could potentially push growth beyond the longstanding two per cent trend, but also suspects the hockey-stick effect remains some time away, as AI is nowhere close to permeating the entire economy.
Jack Dorsey of Block and Roelof Botha of Sequoia have offered a compelling perspective, arguing that layers of middle management exist largely because information can only travel at certain velocities through human communication.
Dorsey and Botha suggest that in an AI-enabled organisation, some kind of AI coordination or “intelligence” system could perform much of the information-routing function traditionally handled by management layers.
Rather than humans spending time passing information up and down a hierarchy, the argument goes that everyone could become closer to being a “doer,” fundamentally altering what an organisation looks like from the inside.
Liu draws a clear conclusion from this line of thinking, suggesting that perhaps middle management, at least in its traditional form, is itself one of the weakest links that AI could ultimately dissolve.
Beyond structural change, Liu also argues that value will accrue to improving human judgement and decision-making, rather than simply automating around human capabilities at the edges of the knowledge economy.
As AI-generated content becomes increasingly ubiquitous, the ability to identify what is genuinely human and what is not will itself become a scarce and therefore valuable quality in any organisation.
The central question, as Liu frames it, is whether humanity can remain part of the economic chain, not merely as the weak link, but as a necessary and irreplaceable one.