Can generative AI replace the benefits of having a human teammate?

Dell'Acqua et al. report, from a pre-registered field experiment with 776 professionals at Procter & Gamble, that generative AI raised the performance of individuals on real product innovation challenges to the level of teams without AI. Professionals were randomly assigned to work with or without AI, and either alone or with another professional in new product development teams. The abstract's headline comparison is that "individuals with AI matched the performance of teams without AI," and the authors conclude that "AI can effectively replicate certain benefits of human collaboration." The same design supports two further claims. AI "breaks down functional silos": without AI, R&D professionals suggested more technical solutions and Commercial professionals more commercially oriented ones, while professionals using AI "produced balanced solutions, regardless of their professional background." And AI's language-based interface prompted more positive self-reported emotional responses, which the authors read as partly filling the social and motivational role a human teammate usually plays.

The excerpt gives the mechanism only in outline. In the team condition, a second person supplies a second perspective, and the authors' reading is that AI supplies something similar to a solo worker, enough to close the gap with a pair. The silo result follows the same logic: when the tool is not tied to one function's habits, the proposals drift toward the middle. The abstract does not say how solutions were scored, what the performance measure was, or how the AI was prompted, so the mechanism is the authors' interpretation of the outcomes, not something the excerpt shows directly.

This finding sits in tension with the nearest notes. The Does generative AI shift knowledge workers away from communication? note reads trace data showing heavy AI users moving away from communication and toward solo document work. Read together, the two suggest that AI can stand in for a human partner in a task while people spend less time on the coordination that partnership involves. The P&G abstract measures output, not time use, so it cannot test that reading. The Does AI assistance help workers learn lasting skills? note reports gains that disappear when workers act alone; this experiment measures gains inside an AI-assisted setting and says nothing about persistence. The Can AI narrow the education performance gap? note is the closest parallel: a randomized result in which AI narrows a background gap. The silo finding makes the same move along functional background instead of education. The Does theory of mind predict who thrives in AI collaboration? note suggests that the benefit of AI partnership depends on the individual, whereas this abstract reports average-style comparisons and does not say whether the gain was uniform across participants.

What the excerpt does not establish is substantial. It is an abstract of 211 words, so effect sizes, test statistics, the performance metric, and the coding of "balanced" solutions are all absent. The emotional finding rests on self-report. The setting is one consumer packaged goods company and one kind of work, new product development, over a period the excerpt does not state. The implication is that the teammate-replacement result is credible as an observation from one firm's innovation tasks, and that the silo result is evidence that AI can shift the perspective of a proposal. Neither supports the authors' broader call for organizations to "rethink the very structure of collaborative work" without the full paper and replication in other settings.