In 2024, researchers from Harvard, Wharton and other institutions ran a large field experiment with 776 professionals at Procter & Gamble working on real product development challenges. One result stands out: individuals working with AI performed just as well as teams without AI.
For any manager, this raises an uncomfortable question. If one person with AI can match a team's output, what is the team actually for?
And it quickly opens others:
How do I keep real collaboration alive when everyone works faster, alone, with their own AI?
Which human skills should I be developing in my team, and how, when AI is always one click away?
And who exactly is in my team, now that some of its “members” are AI agents?
These are questions many managers in AI-transformed organisations are already living with. Few have satisfying answers yet.
The same P&G study offers a more nuanced picture than its headline. Teams using AI were significantly more likely to produce top-tier solutions, suggesting that human collaboration still brings something AI alone does not. The researchers also observed that professionals using AI produced balanced solutions regardless of functional background, whereas without AI, R&D and commercial profiles tended to stay within their own perspective. AI, in other words, can soften silos.
Research from MIT's Centre for Collective Intelligence adds an important caution. In a meta-analysis of 106 experiments published in Nature Human Behaviour, Vaccaro, Almaatouq and Malone found that, on average, human–AI combinations performed worse than the best of humans or AI alone. The picture varied widely, however: combinations tended to lose value in decision-making tasks and gain value in tasks involving creating new content. And when humans outperformed the AI on their own, combining them brought gains; when the AI was better, adding the human caused losses.
For managers, the lesson is not that collaboration with AI fails, but that it is not automatically additive. Deciding who does what, between people and machines and between people themselves, becomes a core part of team design.
There is a second, quieter risk. In a study published in Science Advances, Doshi and Hauser found that AI-assisted stories were rated as more creative, yet they were more similar to each other than stories written by humans alone.
Several executives Solvay Lifelong Learning interviewed this year raised the same concern. When an entire organisation, or even an entire sector, converges on the same AI tools, is the risk a quiet standardisation of thinking? And combined with more asynchronous, individual work, will this erode precisely what made teams valuable: the confrontation of different viewpoints?
AI can also change how connected people feel. In a series of studies published in the Journal of Applied Psychology, Tang and colleagues found that employees who interacted more frequently with AI systems reported more loneliness, which was in turn linked to insomnia and, in some samples, more after-work drinking. The same employees were also more likely to help their colleagues, possibly as a way to reconnect. The findings are correlational, but they echo what many managers observe: a team member who produces more than ever and talks to colleagues less than ever.
Meanwhile, the shape of the team itself is changing. Microsoft's 2025 Work Trend Index, based on a survey of 31,000 workers across 31 countries, describes the emergence of human–agent teams and the “agent boss”, an employee who directs AI agents as well as working with colleagues. It invites organisations to think about their “human–agent ratio”. Whatever you call it, it raises a real question for managers: what does team cohesion mean when part of the work is done by agents that nobody has lunch with?
That's why these questions are hard to solve alone. A manager in an industrial group facing an uneven rollout across divisions will not land on the same answer as a CIO running AI agents in production, or an SME CEO who discovered “shadow AI” before structuring adoption. Yet each can learn a great deal from how the others are reasoning.
Put together, these findings open a set of questions for which there is no textbook answer:
• Which moments genuinely need to be collective, and which can stay individual?
• For which tasks does combining people and AI add value in my team, and for which does it get in the way?
• As AI agents take on parts of the work, how do we define roles and coordination between people and agents?
• How do we prevent hyper-productive individuals from becoming isolated individuals?
• How do we deliberately create productive disagreement when “the AI said so” becomes a tempting answer?
• How do we make sure the shift towards oral and relational skills doesn't create new forms of unfairness within the team?
None of these questions has a single right answer. A tech scale-up running AI agents in production will not answer them like an industrial group with uneven adoption across divisions, or a family-owned multinational several years into its digital journey.
What helps here is not a model to apply, but the opportunity to confront one's reasoning with others facing the same dilemmas in different contexts. Hearing how a CIO organises collective
moments in a team working with AI agents, how an operations director sets norms for AI-assisted work, or how an HR director tackles isolation in hybrid teams, with research to test assumptions and frameworks to structure the discussion, helps managers move from uncertainty to concrete experiments they can launch in their own teams.
This is one of the questions at the heart of Executive Programme in People Leadership in the Age of AI, a new programme starting in Spring 2027 at Solvay Lifelong Learning, designed for managers in organisations already transformed by AI. To help participants find answers that are pragmatic, contextual and actually implementable in their own organisation, the programme combines robust frameworks to read the situation, research to confirm or challenge intuitions, and structured exchange with peers facing similar dilemmas in different contexts, across four in-person modules and small-group online coaching sessions.
In your team, what has changed in the way you collaborate since AI arrived?
Are you interested in joining our first cohort of the Executive Programme in People Leadership in the Age of AI? Find all the information on our programme page.
Dell'Acqua, F., Ayoubi, C., Lifshitz, H., Sadun, R., Mollick, E., Mollick, L., et al. (2025/2026). The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise. NBER Working Paper 33641; published in Organization Science.
Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290.
Tang, P. M., Koopman, J., Mai, K. M., De Cremer, D., Zhang, J. H., Reynders, P., Ng, C. T. S., & Chen, I.-H. (2023). No person is an island: Unpacking the work and after-work consequences of interacting with artificial intelligence. Journal of Applied Psychology.
Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293–2303.