Rapid progress in Artificial Intelligence capability is outpacing real-world adoption, organizational change, and measurable productivity gains. The core policy tension is whether Artificial Intelligence can raise aggregate productivity while distributing benefits broadly across firms and workers rather than concentrating them among dominant companies and owners of key inputs. Diffusion depends not just on access to models, but on complementary assets such as digital infrastructure, management quality, data systems, skills, and governance. Where those complements are weak, adoption tends to remain shallow, and the gains from Artificial Intelligence are more likely to widen gaps between large and small firms, richer and poorer regions, and capital and labor.
Evidence on productivity remains mixed and highly context dependent. In one field study, access to a generative Artificial Intelligence tool increased customer support productivity by an average of 15%, with larger gains for less-experienced workers. A separate trial found that completion time increased by 19% among experienced open-source developers using early 2025 Artificial Intelligence tools, suggesting that complex, context-heavy work can suffer from added cognitive overhead. A February 2026 NBER working paper surveying nearly 6,000 executives across the United States, United Kingdom, Germany, and Australia reports that around 70% of firms “actively use AI,” yet executives’ own usage averages about 1.5 hours per week, and about 90% of firms report no impact on employment or productivity over the prior three years. Macro evidence remains limited, with one exercise estimating an upper bound of about a 0.66% increase in total factor productivity over 10 years, with an average estimate below 0.53%.
Adoption patterns show strong divergence by sector, geography, firm size, and language environment. Eurostat reports that 19.95% of EU firms with 10 or more employees used at least one Artificial Intelligence technology in 2025, while U.K. survey data report about 9% of businesses with 10 or more employees using at least one Artificial Intelligence application in 2023, and private-sector survey results for 2025 estimate 39%. U.S. Census data from February 2026 suggests roughly 17.5% of U.S. businesses used Artificial Intelligence in at least one business function in the last two weeks. English-centric models also create barriers for lower-resource languages, and English accounted for 88% of models with language tags on Hugging Face in 2025. OECD analysis finds leaders are pulling further ahead, not that laggards are catching up.
Small and medium-sized enterprises face particular barriers because they often lack specialist talent, funding, and clear use cases. Structured support can improve uptake: Singapore’s 2025 Digital Economy Report reports a tripling in SME Artificial Intelligence adoption from 4.2% in 2023 to 14.5% in 2024. Governance and public procurement emerge as important policy tools. Frameworks from NIST, ISO/IEC 42001, and the OECD can reduce uncertainty and build trust, while procurement rules can lower vendor lock-in, standardize risk expectations, and open public markets to smaller suppliers. Dialogue participants broadly agreed that meaningful, broad-based Artificial Intelligence diffusion should be treated as a policy objective in its own right, supported by better metrics, complementary investment, labor market transition tools, and more realistic expectations about how productivity gains spread.
