Sources
All accessed 2026-08-19. [vendor] marks vendor-published claims about the publisher's own market.
Dillon, Jaffe, Immorlica and Stanton, Shifting Work Patterns with Generative AI (NBER WP 33795, rev. Nov 2025): the telemetry-measured randomised experiment across 66 firms and 7,137 workers, including the null results and the norm-coordination mechanism; UK Department for Business and Trade M365 Copilot evaluation (Aug 2025): diary study plus observed tasks scored for accuracy and quality; Australian Treasury and Australian Centre for Evaluation Copilot trial (Feb 2025); UK cross-government, DWP and HMRC Copilot evaluations (2025-2026) for the rigor gradient; Humlum and Vestergaard (NBER WP 33777, May 2025) for the administrative-data nulls; Vaccaro, Almaatouq and Malone (Nature Human Behaviour, Oct 2024) preregistered meta-analysis; Dell'Acqua et al., The Cybernetic Teammate (Organization Science, Jul 2026, N=791, industry funding disclosed); Zercher et al. (Journal of Organizational Behavior, Jun 2025) on knowledge distribution; Ye et al. (arXiv 2606.05647, Jun 2026) on oversight failure; Grunde-McLaughlin et al. (arXiv 2602.16844, Feb 2026) on confidence without accuracy; Nixon et al. (arXiv 2607.27179, Jul 2026) on AI conversational dominance and human-to-human relations; Kim et al. (AIES 2026, N=157) on invoked versus ambient architecture; Reif, Larrick and Soll (PNAS, May 2025, N=4,439) on the social penalty; Alsobay et al. (CSCW 2026, N=1,475) and Parisi et al. (arXiv, 2026, N=879) on facilitation nulls, algorithmic steering and the illusion of inclusion; Maier, Schneider and Feuerriegel (2025-2026) on interaction mode and diversity; Microsoft Entra agent identity documentation (2025-2026) [vendor]; Google Workspace Cloud Next announcements (Apr 2026) [vendor]; Netskope cloud and threat telemetry (Jan 2026) and Microsoft/LinkedIn Work Trend Index (2024-2026) [vendor] for shadow AI; Menlo Ventures State of Generative AI in the Enterprise (Dec 2025) for spend composition; ClawsBench (arXiv 2604.05172, Apr 2026) for productivity-agent reliability and unsafe-action rates. Unverified, excluded: assistant seat counts circulating in trade press with no primary disclosure; utilisation percentages attributed to unnamed analyst surveys; a vendor claim about notetaker effects on speech with no disclosed sample; workslop perception percentages (the prevalence, time and cost figures are sourced, the perception splits are not). Re-verify by Q4 2026: the meeting-recording litigation; Facilitator general availability beyond scheduled meetings; agent-identity backfill for agents created before July 2026.
Source: research/R08-productivity-and-collaboration/sources.md in the evidence repository behind this site.