Anthropic CEO Dario Amodei is urging a slower pace of frontier artificial intelligence development for safety reviews, while a global survey shows adoption expanding faster than reported companywide earnings gains.
Consulting firm McKinsey surveyed 1,719 people in 97 nations from May 4 to June 8, 2026. It found that 44% said their organizations were scaling AI across the enterprise, up from 38% a year earlier. The share who credited AI with a positive effect on earnings before interest and taxes (EBIT) was 37%, essentially unchanged from the previous year.
McKinsey classified 6% as AI high performers: respondents who attributed at least 5% of their organization’s EBIT to AI and reported significant value from it. A much larger share, 80%, said AI had improved their personal productivity.
In a September essay, Amodei argued that commercial incentives could push advances in model capability ahead of safety work. He said Anthropic had committed to embedding outside evaluators. His broader proposal for shared safety standards and a slower development pace would require cooperation from other companies and governments. He did not present slower development as a way to raise corporate earnings.
Company practices may help explain differences in reported results. Nearly three-quarters of McKinsey’s high performers said they had substantially redesigned workflows, compared with about one-quarter of other respondents. The survey does not establish that those changes caused higher returns.
Adoption also varied by company size. Among respondents at businesses with more than $1 billion in annual revenue, 40% reported scaling AI agents in at least one function, up from 27% a year earlier. The figure at smaller companies was 22%.
The transition carries costs for businesses and workers. About 20% of respondents said AI operating expenses constrained use. Another 32% said their organizations had decided against buying a software product or feature because they could build the functionality internally with agentic coding tools. Among respondents at organizations using AI, 14% reported an AI-related workforce decline over the previous year. Separately, 39% of respondents expected a decline in the following year.
Task-level research offers a narrower view of productivity. In a randomized METR study, 16 experienced open-source developers completed 246 tasks from February to June 2025 and took 19% longer when AI tools were available. METR said in a Feb. 24, 2026, update that later results suggested possible speedups, but participant and task selection prevented a reliable estimate of their size. Neither that research nor McKinsey’s survey tests how a coordinated slowdown would affect company earnings.
