The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork
Citation. Dell’Acqua, F., Ayoubi, C., Lifshitz, H., Sadun, R., Mollick, E., Mollick, L., Han, Y., Goldman, J., Nair, H., Taub, S., & Lakhani, K. R. (2026). The cybernetic teammate: A field experiment on generative AI and teamwork. Organization Science, 37(4), 1217–1242. https://doi.org/10.1287/orsc.2025.20702
Summary. This peer-reviewed, preregistered field experiment randomly assigned 791 Procter & Gamble professionals working on real product-innovation challenges to work individually or in cross-functional pairs, with or without generative AI. AI-supported individuals matched the solution quality of non-AI teams, and AI access reduced functional specialization in ideas and improved short-run self-reported affect. The published analysis further locates the performance benefit in generating higher-quality ideas rather than in selecting among them, while retaining human evaluative judgment as a distinct function.
Key claims
- In the main solution-quality analysis (550 observations), teams without AI improved by 0.245 standard deviations over individuals without AI, individuals with AI by 0.373, and teams with AI by 0.392; AI-supported individuals were statistically comparable to non-AI teams.
- Without AI, R&D professionals produced more technically oriented ideas and Commercial professionals more commercially oriented ideas; AI-supported professionals produced more balanced solutions regardless of functional background.
- AI-supported individuals increased positive emotions by 0.457 standard deviations and AI-supported teams by 0.635, while both AI conditions reduced negative emotions by 0.233–0.235 standard deviations relative to the control condition.
- Decomposing innovation performance, the authors find that AI primarily improved the quality of generated ideas rather than selection accuracy, so human judgment remains important in evaluative selection.
- AI-assisted solutions showed greater semantic similarity in embedding space, raising an unresolved question about whether higher mean idea quality can coexist with reduced diversity in an organization’s innovation portfolio.
Evidence & limitations
- Design: preregistered 2 × 2 field experiment in a one-day virtual P&G product-development workshop; 791 randomized participants completed the workshop and 776 provided complete post-task survey data. The main solution-quality analysis used 550 observations, with blinded expert evaluation of outputs.
- Evidence strength: peer-reviewed causal evidence in a real organization and business domain on short-run idea quality, functional breadth, and self-reported affect.
- Limitations: one company, one early-stage product-innovation task family, and a short intervention; the study does not establish long-run learning, implementation quality, coordination, trust, team development, staffing effects, or equilibrium labor substitution.
- Version/provenance: this record replaces the 2025 HBS working-paper/SSRN version (DOI 10.2139/ssrn.5207588). The published Organization Science PDF has a different binary hash, page count, title, and analysis than the prior 56-page working paper; its DOI and Crossref metadata identify the canonical peer-reviewed version.