HBS AI Institute — Think you’re a good leader? Try leading an AI team
TL;DR
A ~970-word research summary by the HBS AI Institute’s content team, published 3 September 2026, of Weidmann, Xu & Deming (2025). Deming is a 2026 associate of the institute. It restates the paper in three “key insights” and adds one of its own:
- One test, two kinds of teammates. 249 leaders each led six human teams and six teams of GPT-4o agents on Hidden Profile puzzles.
- The leader, not the luck of the draw. Leader identity explains more than half the variation in team performance; AI-test and human-test scores correlate 0.67 raw, 0.81 corrected; a good leader solves 53% of problems, a weak one 10%.
- Successful leaders ask, listen, and reassess. “If you assume the best leaders are the ones who talk the most, the data politely disagrees.” Questions, turn-taking and “we” predict success; warmth matters more with human followers.
- The institute’s own addition: AI as a leadership practice environment. “Managers can rehearse questioning, delegation, information gathering, and decision-making with AI teams, further enhanced by employing a range of personas. Instead of treating leadership development as an occasional workshop, leaders and organizations can build short, repeatable simulations around familiar challenges.” It adds a caution: simulations suit “the mechanics of coordination,” but development must still address “the trust, morale, and emotional judgment that shape real human teams.”
How it relates to the paper
The figures are reported accurately. Two differences in emphasis:
- Use case. The paper argues for measurement: selecting leaders, evaluating training, cheaper research. The summary argues for practice: rehearsing leadership with AI teams. The paper does not test whether practising with AI agents improves leadership; that is the summary’s extrapolation.
- What is left out. The summary omits the authors’ main stated limitation, that the AI test has not been validated against leadership success in real organisations, and the cost comparison ($114 vs $23 per participant).
It also misspells co-author Yixian Xu as “Yixian Yu” in the author bios.
Dynamic-capabilities reading
digital-transforming/improving-digital-maturity— the practice-environment proposal is a workforce-capability lever: repeatable AI-team simulations as a way to develop leadership, including the leadership of AI agents themselves.
How it connects
- durable-skills — the practice-environment idea extends the paper’s measurement claim to development; see the paper’s page for the evidence.
- Mastering Change Resilience — linked from the article’s “bonus” section as the organisation-level counterpart.
Linked entities and concepts
- HBS AI Institute — publisher. Entity created in this ingest.
- durable-skills — updated in this ingest.
Dangling: see the paper’s page for the authors.
Scope and reliability
A secondary source: an institute’s summary of a paper by one of its own associates. Cite the paper for any finding; cite this page only for the practice-environment proposal, which is the institute’s opinion and untested.