Managers accepted reflective coaching from humans but wanted answers from AI

Eight managers welcomed probing questions from a human coach but resisted the same approach from a text-based AI coach.

Editorial illustration — not from the study.

Researchers adapted a leadership-coaching protocol previously supported in earlier cycles into a text-based AI coach. In the fourth cycle of a design-based research program at a global consulting firm, eight managers used the protocol with both a human coach and the AI coach during observed onboarding activities, then discussed their experiences in interviews.

All eight participants welcomed probing questions from the human coach and preferred the human coach overall. They resisted the same questioning approach from the AI coach and instead wanted the system to provide answers. The researchers describe this mismatch as a “Reflection Penalty” and propose dividing coaching tasks between humans and AI according to which is better suited to each job.

How managers responded

All eight managers welcomed probing questions from the human coach and preferred that coach. They resisted receiving the same reflective questioning protocol from the AI coach and wanted answers from the system instead. The researchers argue that an AI coach may face a “Reflection Penalty” when it asks users to do the reflective work that helps make human coaching effective. Participants also suggested different tasks for human and AI coaches, which the researchers organized into a jobs-to-be-done map.

Why the mismatch matters

Leadership coaching can help develop managers but is expensive and difficult to scale. The findings suggest that simply transferring a human coaching protocol to an AI system may not produce the same experience: users may accept reflective questioning from a person while expecting an AI system to supply information or answers. The proposed task map offers one way for designers to decide which parts of coaching should be handled by humans and which could be assigned to AI.

Evidence and limits

The study was the fourth cycle of a design-based research program at a global consulting firm. It involved eight managers, who used the same protocol with a human coach and a text-based AI coach, applied each session in an observed onboarding activity, and discussed both experiences in interviews. The small group and single organizational setting limit how broadly the findings can be generalized. The abstract also does not report quantitative measures of learning or compare coaching outcomes beyond participants’ reported preferences and observed activities.

Tuna and billfish fishing moved offshore mainly in the past 200 years

People have fished for tuna for about 44,000 years and billfish for about 10,000 years, according to a global review of archaeological evidence. But sustained, intensive fishing in offshore and oceanic waters appears to have developed mainly during the past 200 years.

5 hours ago·3 views

Abstract editorial illustration accompanying the article: Coastal residents weigh flood damage, costs and family ties when considering a move

Society & Economics

Coastal residents weigh flood damage, costs and family ties when considering a move

People considering whether to leave flood-prone coastal communities weigh disastrous flooding and rising costs alongside crime, healthcare and proximity to family and friends. In a survey of 182 respondents, most said they would not relocate, while others said they would move immediately or in the future.

13 hours ago·5 views

Abstract editorial illustration accompanying the article: AI voter personas shifted an election estimate toward Democrats—but accuracy is unclear

Society & Economics

AI voter personas shifted an election estimate toward Democrats—but accuracy is unclear

Researchers used large language models to create artificial survey respondents for a weekly look at the 2026 US House elections. Adding personal memories and news exposure shifted the estimates toward Democrats, but the resulting 18.7-point Democratic margin was well above the 8.7-point polling average, leaving the method’s predictive accuracy unresolved.

1 day ago·37 views