The exact NBER figure varies by source. Some write-ups report 89% of managers rather than 90% of executives. Several of the numbers in the piece also come from secondary sources, including vendor blogs and summaries, not the original papers. Those papers should be checked before any figure is repeated. The piece also cites a 2025 METR randomized trial in which experienced developers using AI tools finished 19% slower while believing they were 20% faster. Stanford's 2026 AI Index reports task-level gains of 14 to 15% in customer support, 26% in software development, and 73% in marketing output. The gap between those task results and firm-level outcomes is the central unresolved question. A claim that over 40% of agentic AI initiatives will be cancelled by 2027, attributed to Gartner, appears only in some sources.
For attorneys, the divergence matters in practice. A WSJ analysis cited in 2026 found that 40% of workers report saving no time at all using AI, while executives are far more optimistic. Any deal diligence, vendor representation, or client advice that treats AI productivity gains as established should separate task-level evidence from company-wide results. Workforce decisions justified by AI efficiency deserve the same scrutiny, since the executive-employee gap is where disputes are likely to surface.