Pair people with AI for a decision and the team often does worse than the smarter partner alone — but for creative work, the pairing wins.

This explores why human-AI pairs tend to do worse than the stronger partner alone when making decisions, yet do better when making things. The pattern comes from a meta-analysis of more than 100 studies. On average, human-AI combinations scored below the best of human or AI alone, but content-creation tasks were the exception, where real gains showed up When do human-AI teams actually outperform either alone?. The authors put the difference down to how the tasks are structured, not to what either partner can do. That turns out to be the key to the whole question.

Think about what a decision task asks for: one right answer. If the AI is better, the best a human can add is knowing exactly when to overrule it, and people are bad at that. A framework that splits teamwork into two sources of gain makes this clear Can humans and AI teams achieve performance neither reaches alone?. The first is 'inherent' potential: picking whoever is right on each case. The second is 'collaborative' potential: actually combining reasoning. Decision tasks lean almost entirely on the first, and it only pays off if someone can reliably judge which partner to trust on each case. Usually nobody can, so the team settles somewhere between the two and lands below the better one. Even a simple ensemble beats both partners only when their mistakes are genuinely different from each other's Can humans and AI improve decisions by predicting together?. In practice, humans who defer to AI end up making the AI's mistakes too.

Creation is a different shape of problem. There is no single right answer to land on, so contributions add up instead of crowding each other out. A human's angle and an AI's breadth can both go into the final piece. In a field experiment at Procter & Gamble, individuals working with AI produced solutions as strong as two-person teams without it. They also produced more balanced ideas across professional specialties, as if the AI stood in for the colleague from another department Can generative AI replace the benefits of having a human teammate?. Research on AI teams generating ideas adds a warning: variety of viewpoints only helps when real expertise sits underneath it. Diverse teams without expertise do worse than one competent member working alone Does cognitive diversity alone improve multi-agent ideation quality?.

The less obvious lesson is that the decision-task losses look like an organization-design failure, not a flaw in the technology. Puranam treats a human-AI pair as a small organization, where results depend on how work is divided and how each side gets feedback and learns How should humans and AI systems divide decision-making tasks?. Agrawal, Gans and Goldfarb show that AI can improve each individual decision while making decisions less consistent, which hurts whenever decisions depend on each other Does AI adoption require rethinking how organizations coordinate?. The teaming research agenda points the same way: teams need clearly specified roles that can still shift as the work changes, plus calibrated trust What blocks effective human-AI teams from performing well?. Most decision-task studies just put a human next to an AI recommendation and ask them to approve or reject it, which is the weakest arrangement possible.

So 'AI teams underperform at decisions' really means: dropping AI advice into a human's workflow, with no thought given to roles, feedback, or how the two partners' errors differ, usually backfires. Creative work hides that design problem because its contributions add up naturally. Decision work exposes it.