The pitch has always been democratic: AI will help everyone, lower the barriers to expertise, and give a junior analyst the capabilities of a senior one. MIT economist and Nobel laureate Daron Acemoglu is not buying it. A new survey and Acemoglu’s analysis, reported this week by the Financial Times, make a blunter case — AI is not leveling the playing field; it is tilting it further toward those who already hold the strongest hand.
Who Actually Benefits
The core finding is about access, not theory. “The rhetoric out there is that the tools are going to be democratizing,” Acemoglu told the Financial Times. “But the reality is that you require a certain degree of education, abstract and quantitative skills, familiarity with computers and coding in order to be using the models.” In other words, the workers best positioned to leverage AI are already the most economically secure — knowledge workers at large firms, software engineers, financial analysts. The assembly-line worker, the call-center agent, the data-entry clerk: their roles are being automated, not augmented.
The survey data backs the argument. Across surveyed firms, workers reporting productivity gains from AI tools skewed heavily toward those with college or postgraduate degrees and roles involving significant computer-based work. Workers in lower-wage, task-based roles were more likely to report that AI tools had reduced their workload — by eliminating their position.
Capital vs. Labor: The Structural Divide
Acemoglu’s framing cuts deeper than individual job displacement. His concern is systemic: AI’s productivity gains are captured by capital owners — the shareholders of companies deploying the technology — rather than by the workers whose efficiency it improves. “AI is going to increase inequality between labour and capital. That is almost for sure,” he said. “I would say it is setting us up for a shitshow” (Financial Times, April 2026).
The macro signals support his concern. Google CEO Sundar Pichai disclosed this week that 75 percent of all new code at Google is now AI-generated, up from 50 percent last fall. Anthropic’s own disclosures suggest Claude Code handles 70 to 90 percent of code written internally. Neither announcement came with proportional hiring reductions — but neither came with hiring expansions either. The productivity gains are being captured as margin, not redistributed as wages.
The broader structural issue is that AI investment is running far ahead of any policy framework capable of managing its distributional effects. The EU AI Act focuses on safety and risk categories; it does not address labor market dynamics. In the United States, no federal framework exists. Tax policy has not adjusted for a world in which machines generate measurable GDP-equivalent output.
The Policy Gap
Acemoglu has been consistent on the need for intervention: a reorientation of AI development toward tools that genuinely complement workers rather than substitute for them, combined with tax treatment that reduces the current incentive to automate rather than hire. Neither shift appears imminent.
What is imminent is the continued acceleration of capability deployment into enterprise workflows. Anthropic’s app connectors, Microsoft’s Copilot integration into every M365 seat, and Google’s Gemini embedding across Workspace are all, in aggregate, a rapid restructuring of white-collar work. The question Acemoglu is raising is not whether this transformation will happen — it clearly will — but who bears the cost of it.
The honest answer, based on current trajectory: not the shareholders.
Sources: Financial Times / Daron Acemoglu interview, The Verge (April 23, 2026), Google Cloud Next blog, Anthropic disclosures
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.