AI's Productivity Promise: What the Research Actually Says About Impact and Timelines

AI promises to boost productivity dramatically—but what does the evidence actually show? Oriol Carreras Baquer of CaixaBank Research offers a sober, research-grounded assessment: the gains are real, but the path from individual task improvements to economy-wide transformation is longer and more uncertain than the hype suggests.

Individual productivity gains from AI tools are striking. A review by the OECD finds that AI use can boost worker productivity by around 30% on average, with some studies showing improvements above 50% in specific tasks like programming and writing. But these studies test AI under favorable conditions—they don’t account for training costs, process redesign, or organizational friction.

Key Takeaways

• The macro impact depends on how many tasks AI actually changes. Economist Daron Acemoglu estimates that only about 5% of tasks will be economically viable to automate within the next decade, suggesting modest GDP gains of ~0.1% per year. More optimistic OECD models project annual productivity gains of 0.4–1.3 percentage points in the US over 10 years—significant, but not overnight.
• AI acts as a “leveller” within job categories. Across many tasks, AI tends to help lower-performing workers the most, narrowing performance gaps rather than only empowering top performers.
• Employment effects depend on which force wins. Automation displaces tasks. But new technology historically creates new jobs too (the “reinstatement effect”). Whether AI replicates that pattern—and how fast—is the defining open question for the labor market.

The most likely scenario: productivity gains grow meaningfully in the medium term as companies redesign entire processes—not just individual tasks—with the US moving faster than Europe given its higher pace of tech adoption.

Read the full article on CaixaBank Research.