It seems plausible that the developing AI technologies will improve productivity. I have friends and family who tell me how it is already useful in their own everyday work, often for speeding up the process of producing first drafts: say, of a Powerpoint slide for display, or a recommendation for an internship, or the software programming to solve a problem. A common theme, it seems to me, is that the current form of AI tools can save considerable time at the earlier stages of a task, but at least for these kinds of tasks, the AI tools almost never get you close to a final version with the level of quality desired.
What about how AI might affect productivity growth for the economy as a whole? A group of OECD economists–Francesco Filippucci, Peter Gal, Katharina Laengle, Matthias Schief, and Filiz Unsal–offer an overview and discussion of existing estimates in “Opportunities and Risks of Artificial Intelligence for Productivity” (International Productivity Monitor, Spring 2025). Here’s their bottom line:
AI could contribute between 0.3 and 0.7 percentage points to annual aggregate
TFP growth in the United States over the next decade. The predicted impacts across different scenarios are highest in the United States, followed by the United Kingdom, Germany, Canada, France and Italy, and lowest in Japan. These figures indicate that Generative AI will likely be an important source of aggregate productivity growth over the next 10 years but also clarify that the expected gains from the current generation of AI technologies may not be extraordinary. For comparison, the latest technology driven boom linked to information and communication technologies (ICT) has been estimated to have contributed up to 1-1.5 percentage points to annual TFP growth in the United States during the decade starting in the mid-1990s …
This seems to be a mid-range forecast, compared to other published estimates. Here’s a figure showing the gains from use of AI in some specific studies for specific tasks. The vertical axis can be read as “%”– that is, AI tools improved customer service by 14%, improved coding by 56%, and so on.
How fast will AI tools be adopted? Here a figure showing roughly how long it took other systematically important technologies to be adopted by half of firms:
As these figures suggest, a range of factors can lead to higher or lower estimates for the productivity effects of AI. Here are a few of them:
- At what rate will AI tools be adopted across the economy?
- To what extent will AI speed up existing tasks for what share of current workers and jobs?
- To what extent will AI foster skill development for existing workers, allowing them to carry out additional tasks?
- Will making full use of AI tools require complementary investments in skills or physical capital that will take time and funding before they are implemented?
- To what extent will AI speed up research and innovation?
- Will the AI tools and the underlying technologies that allow use of those tools (like high-speed connectivity, computing power, and specialized computer chips) be made broadly available in a competitive market that tends to drive down the prices paid by users, or will they be sold in a less competitive environment with higher prices–so that they are used less frequently?
- To what extent will AI tools allow firms to exploit behavioral biases or malicious activities, in a way that results in lower overall gains for society?
The answers to these questions (and others) clearly open up an array of possibilities. As one final concern, the authors point out that “[h]istorically, sectors experiencing faster productivity growth have in fact tended to see decreases in their GDP shares (driven by declines in relative output prices and employment shares), thus reducing aggregate productivity growth …” Total productivity growth for the economy as a whole, of course, must include both sectors that make effective use of AI tools, along with sectors that don’t. The gains from AI in specific jobs and tasks is thus only one slice of the productivity picture.
