Technical essay
After adoption
Workplace GenAI is widespread enough that adoption alone no longer describes the workday. The next measurement problem is conversion: how reported adoption turns into recurring use, assisted working time, reported savings, and realized economic outcomes.
2026-Q2
2026-Q2
2026-Q2
In 2026-Q2, 39.2% of employed U.S. adults aged 18–64 reported work use in the prior week and 13.7% reported use on every workday. The three headline measures above describe reach, active assistance, and reported time released.
From 2025-Q2 to 2026-Q2, work adoption rose 28.8%, weekly use rose 28.4%, assisted hours rose 27.8%, and reported savings rose 19.6%. Their rates of movement differ across reach, recurrence, use intensity, and reported time released.
01 · Reach and depth
Diffusion has a more stable map than workflow penetration.
Adoption counts workers who report using GenAI for work. Recurrence and time intensity answer different questions. Across 7 common subgroup quarters from 2024-Q4 through2026-Q2, the cross-group ordering of adoption is more persistent than the ordering of assisted working time.
Adoption rankings exceed assisted-hours rank stability in 20 of 21 industry quarter-pair comparisons and 20 of 21 occupation comparisons.
02 · Occupation structure
Occupation organizes conversion more tightly.
In every audited quarter, both Pearson and Spearman correlations between adoption and assisted hours are higher across occupations than across industries. At the occupation level, adoption also has the higher univariate R² for reported savings in 7 of 7 waves and all 154 leave-one-occupation checks.
The repeated ordering makes occupation the strongest organizing layer in the current aggregate evidence. Industry-level relationships are less stable: assisted hours has the higher univariate R² for reported savings in 4 of 7 waves, with the ordering changing across the audited window.
03 · Industry context
Occupation mix explains one layer of industry variation.
CPS occupation weights generate an industry counterfactual from national occupation-specific GenAI rates. Adoption uses worker-share weights; assisted hours and reported savings use actual-main-job-hour weights. Observed industry values minus these counterfactuals produce an occupation-adjusted industry-context residual.
Release 1 treats that residual as a descriptive diagnostic. Organizational effects, management quality, efficiency, and productivity require separate identification. Full design-based uncertainty for the custom pooled CPS composition vectors remains unsupported. OEWS supplies an independent establishment-side robustness check.
04 · Interpretation
Reported savings stop short of productivity.
RPS users estimate the additional hours the same work would have required without GenAI. The resulting measure is a self-reported counterfactual estimate of released labor input. Productivity requires observed output or value added plus an identification strategy linking changes in AI use to changes in production.
Adoption remains the first coordinate of workplace AI. Depth is the next: recurrence, assisted working time, reported time released, and eventually measured outcomes. Keeping those stages separate makes the observatory useful as the evidence expands.
Numeric longitudinal claims on this page come from the same versioned derived diagnostic artifact used by the explorers. Public rendering uses release-bound evidence and excludes private source-input history.