A 349-person METR survey of technical workers (software engineers, researchers, academics, founders) conducted February–April 2026 finds median self-reported value uplift of 1.4–2x from AI tools, with a retrospective 1.3x for March 2025 and a forecast 2.5x for March 2027; the median self-reported speed uplift is higher at 3x. The authors explicitly caution that surveys systematically overstate gains relative to controlled experiments, their own prior RCT found a 40-percentage-point overestimation, and qualitative review of the highest-uplift respondents confirms visible overstating in at least 2 of 7 cases. For financial institutions, the data provides a directional (not quantitative) signal that perceived productivity gains among AI-heavy technical workers are rising year-on-year, but the gap between survey perception and experimental reality remains unresolved and material for ROI underwriting.
310 other tracked developments touching the same claims.