paul krugman has a line: "productivity isn't everything, but in the long run it's almost everything." this is one of those statements that sounds like an economist being boring and is actually a statement that, if you really absorb it, will change how you look at everything.
here is the math. a country growing productivity at 2% annually doubles its living standards every 35 years. a country growing at 1.4% annually doubles its living standards every 50 years. that gap — 0.6 percentage points — is not a rounding error. that's fifteen extra years before wages double. that's a generation of workers who got paid less than they would have, not because anyone was cruel to them, not because of any particular policy failure, but because the slow-moving productivity variable had settled 0.6 points lower than it might have. the US experienced exactly this: 2% growth from 1947 to 1973, 1.4% from 1973 to 1995. economists still argue about why. (oil shock, compositional shifts in the economy, underinvestment in infrastructure, various candidates.) workers felt it in their paychecks without being able to name what was happening.
meanwhile the political debates that consume most of our attention — tax rates, trade deals, minimum wage, immigration — are all plausibly second-order effects compared to whether productivity grows or doesn't. this is not to say they don't matter. they do. but over fifty-year horizons, they're small perturbations on the path of living standards compared to the productivity trend itself. you can win every argument about the minimum wage and lose on productivity and the workers you were trying to help will still have lower real wages than if you'd lost every minimum wage argument but found two extra points of productivity growth somewhere. the distribution problem is real and the size-of-the-pie problem is also real and they get about a 20:1 ratio of political attention that is probably the wrong ratio.
the frustrating part is that we don't have a reliable recipe for accelerating productivity growth. we know some ingredients: R&D investment, education quality, good institutions, competitive markets, general-purpose technologies that diffuse widely. but the big jumps are hard to engineer and often impossible to predict. nobody sat down and said "containerized shipping will be the productivity catalyst of the postwar era." it was a trucking company owner (malcolm mclean) who couldn't afford to unload his cargo and wondered if the whole truck body could be lifted onto a ship. electrification took forty years to fully diffuse through American manufacturing. robert solow said in 1987 that "you can see the computer age everywhere except in the productivity statistics." he was right: the productivity boom from computing showed up in the late 1990s, a decade after his quip, as all the organizational and operational restructuring that the technology enabled finally happened.
this lag is important. general purpose technologies — electricity, computers, the internet, potentially AI — don't produce productivity gains immediately because adopting a new technology is not enough. you have to reorganize around it. the classic study is of electrification in factories: factories that adopted electrical motors in the 1880s and 1890s didn't see big productivity gains initially, because they just replaced the central steam engine with an electric motor and kept the same hub-and-spoke factory layout designed around the constraints of the old technology. the productivity gains arrived later, when factory architects redesigned plants from scratch assuming electrical power could be delivered anywhere, and the whole industrial logic changed. the technology came first; the reorganization, which is where the productivity actually lives, came twenty or thirty years later.
all of which makes me somewhat skeptical of confident near-term claims about AI-driven productivity booms. not because i doubt that AI is transformative — i think it probably is — but because the pattern of general purpose technologies suggests the productivity statistics will lag the technology by years or decades while organizations figure out what restructuring the technology actually enables. we're in the "replace the steam engine with an electric motor" phase. the "redesign the factory" phase is coming, and that's where the numbers will move, and it will take longer than the hype cycle suggests.
for the builders: the software that actually drives productivity growth is usually boring. databases. compilers. spreadsheets. version control. containerization. the unsexy infrastructure that lets people do real work faster. a lot of software makes people feel productive without making them productive — notifications, dashboards, activity feeds, engagement mechanics. the question worth asking for anything you build is: does this help people produce more output, or does it just make them feel busy? these are different, and only one of them compounds into the macroeconomic statistics that determine whether wages double in 35 years or 50.