
This episode examines how AI that flatters users—"sycophantic" AI—can reduce idea diversity, foster over‑reliance, and create groupthink. A randomized trial at BT Group showed an AI designed to challenge employees produced 3.5× more excellent goals than default AI or no AI, without harming psychological safety.
Key takeaways: AI design shapes culture; agreement feels good but isn’t progress; human skills like constructive feedback are more important than ever. Practical steps include asking uncomfortable questions, changing prompts to solicit critiques, measuring outcomes (not just usage), testing interaction styles before scaling, and adding human checkpoints to decisions.
More episodes from "The Manager Lab"



Don't miss an episode of “The Manager Lab” and subscribe to it in the GetPodcast app.








