The Reflection Penalty: when managers accept leadership coaching from a human but not an AI agent

Abstract

Leadership coaching develops managers but is expensive and difficult to scale. AI may be able to help scale access, but “answering machine” like AI coaching agents invite cognitive offloading and weaken learning. We instantiated a human coaching protocol, supported across prior cycles, as a text-based AI coach built to prompt reflective questioning. In the fourth cycle of a design-based research program at a global consulting firm, eight managers received the same protocol from a human coach and the AI coach, applied each session in an observed onboarding activity, and reflected on both in interviews. All eight welcomed the probing questions and preferred the human coach. Notably, they resisted the same protocol of questioning from the AI coach and wanted answers from it instead. This study’s theoretical contribution is the Reflection Penalty, the cost an AI coach pays for asking a user to do the reflective work that makes human coaching effective. Its practical contribution is a jobs-to-be-done map for AI coaching, assembled from proposals participants raised on their own. It divides coaching into distinct jobs and assigns each to the human or the AI best suited to it. Both contributions are offered to those designing AI tools that aim to support reflection