In 2011, an essay by venture capitalist Marc Andreessen captured the tech world’s imagination. Titled “Why software is eating the world,” it argued that digital-first companies like Netflix, Amazon and Spotify were poised to remake the American economy — an impressive feat of shot-calling that helped cement Andreessen (a Web business pioneer and recently minted Trump donor) as one of Silicon Valley’s thought leaders. Now in a new working paper from the National Bureau of Economic Research, economists Sangmin Aum and Yongseok Shin take a deep look at the real economic impact of software and make an equally provocative point: that how software eats the world is what really matters. They argue that software works differently from other big economic disruptors. In this case, it really does replace humans — or, in economic terms, it directly allocates income away from workers to the corporations buying software en masse. This is crucially important to policymakers and governments struggling to think about the coming impact of artificial intelligence. Shin told me that with AI poised to remake how software itself is developed, sold and implemented, this trend could outstrip global policymakers’ still very nascent efforts to prepare the world for the future digital economy. “Technological advances are on the surface a good thing,” Shin said. “But at this level they have a very disruptive effect and very big distributional consequences … some people are fortunate enough to take advantage of it, and others see their career opportunities dwindle.” The researchers wanted to solve a puzzle: why workers’ share of total income in the economy has been steadily declining since the 1980s. While the economy has grown to once-unthinkable proportions since then, workers’ earning power has not. That gap has had profound political consequences, inspiring critiques from Bernie Sanders on the left to JD Vance on the right. Shin and Aum’s paper argues that software is directly responsible. Their findings are stark, especially for white-collar workers: In economic terms, old-school machines actually improved productivity for workers. Software, however, flat-out replaces them. They discovered this by studying data from South Korea, where firms track their investment in software independently of their investment in other forms of capital. They found that companies that invested heavily in software saw big productivity boosts when software became cheaper, and that those firms “tend to have lower-than-average labor shares.” “What’s really happening is that firms, instead of hiring five secretaries, now hire only two secretaries who are knowledgeable and who can use software very well,” Shin said. “For white-collar workers over the last 25 years or so there’s been a subtle shift, demanding more investment in software skills on the side of workers while the gains in productivity don’t go back to them.” AI will only accelerate this phenomenon, he says. While the paper’s data predates the current boom in generative AI, the authors point to another recent NBER paper that shows that it serves a similar function as a “substitute” for labor in the economy. “It will be very high-skilled workers, a much smaller segment of workers who can really take full advantage of generative AI,” Shin said. Writing in Bloomberg earlier this year on the overall phenomenon of workers losing out as global profits expand, the economist Tyler Cowen voiced the same idea about AI while proposing another software-related explanation. Cowen wrote that it’s become so easy to be an entrepreneur, in part because of software, that more people are entering the “capitalist” class rather than working as pure laborers, something Shin finds plausible but unlikely to explain more than a small part of the phenomenon. In that case, what can policymakers actually do about it? There, it gets a little fuzzy. The Biden administration has claimed to empower workers by subsidizing the manufacturing industry, among other tentpoles of its post-neoliberal “Bidenomics” agenda, but software isn’t going away anytime soon. In his acceptance speech at the Republican National Convention Wednesday night, vice presidential pick Vance vowed to “commit to the working man” and break from decades of GOP orthodoxy around the purported benefits of ever-increasing corporate profit, but it remains unclear what a governing Republican Party would actually be willing to do to redistribute the economic pie. The researchers themselves say it might not be as easy as passing a law, or restaffing the National Labor Relations Board. Shin described the impact of software on the economy as so large and pervasive that it eludes any one policy solution, and requires a comprehensive investment in education to ensure that workers can interact effectively with the sophisticated AI systems set to gobble up more of the economy’s value. “It’s really difficult to retrain workers, and education is a slow-moving field,” Shin said. “We thought that teaching coding was the best thing we could do, but now that’s unclear … I don't think there's a really simple policy recommendation, other than to have a good education system.”
| 
