Artificial intelligence is driving a surge in productivity and wages, but that growth may be nearing its peak, according to Ioana Marinescu, an associate professor at the University of Pennsylvania’s School of Social Policy & Practice. In a new paper co-authored with Konrad Kording and published by the Brookings Institution, Marinescu introduces the concept of "intelligence saturation" — a model suggesting that as AI automates more cognitive tasks, wage gains will follow a hump-shaped pattern: rising initially, then leveling off, and eventually declining if automation continues unchecked. Marinescu estimates that over 14% of intelligence-based work is already automated, a figure derived from data showing a drop in routine cognitive jobs from 49% in the 1970s and 1980s to 35% by 2018. She warns this shift is already close to the point where wage suppression could begin. In the Brookings model, wages start to fall when roughly 37% of intelligence tasks are automated — a threshold that could be reached sooner than expected if AI adoption accelerates. So far, there’s no broad evidence of a wage decline, but Marinescu points to early warning signs. A recent Stanford study found that early-career workers aged 22 to 25 in AI-exposed fields like software development and customer service have seen a 13% drop in employment since the rise of generative AI, while older and less-exposed workers have seen stable or increasing job numbers. The true red flag, she says, would be a measurable reduction in the overall share of intelligence jobs across the economy — a sign that the labor market is shifting toward more physical, less automatable work. However, a wage downturn is not inevitable. Marinescu argues that the outcome depends on how society manages the transition between the physical and intelligence sectors. She views these two areas as complementary — much like labor and capital — where both are needed to drive production. When AI and human work enhance each other, productivity and wages can continue to grow. But if AI is deployed without investment in human roles, the long-term benefits of automation will eventually diminish. The Brookings paper recommends slowing the pace of automation in cognitive work and increasing investment in physical capital — such as infrastructure, manufacturing, and healthcare — to ensure that human labor remains essential. It also proposes taxing virtual labor, a policy idea similar to Sen. Bernie Sanders’ “robot tax,” to discourage companies from replacing workers with AI and to incentivize a balanced economy. The key uncertainty, Marinescu says, is how substitutable AI and human output become. If AI can fully replace cognitive work, wage growth will likely stall. But if humans and machines continue to work together, each amplifying the other’s strengths, the economy can keep growing without leaving workers behind. The path forward, she argues, isn’t to slow AI, but to shape its integration with smart, strategic investment in the real-world economy.