Khan Academy embeds Khanmigo in practice, founder says first version fell short

Khan Academy founder Sal Khan wrote on July 15 that the first version of Khanmigo, launched three years ago, did not change student learning as much as expected, while rejecting claims that the AI tutor had failed or been abandoned. The revised Khanmigo is now integrated into Khan Academy's main practice experience, can see the problem a student is working on, asks them to explain their reasoning, and uses information about recent performance and skill progress. Chief Learning Officer Kristen Eignor DiCerbo said learners show improvement on similar problems attempted without Khanmigo after interactions, but no figures, sample size or methodology were provided.

This is a public correction by a leading nonprofit education provider about the early results of an AI tutor, showing that layering general conversational ability onto course content does not necessarily change learning behavior. For education product teams, the signal is to embed AI in existing practice flows and drive interaction from the student's current problem and skill data rather than creating a separate entry point; for schools and teachers, it positions AI tools closer to in-class immediate feedback than a replacement for instruction. Notably, neither Khan nor DiCerbo released sample size, research design or effect size, and the math gains in Newark Public Schools were attributed to Khan Academy overall rather than Khanmigo specifically, so the available evidence is insufficient to judge the new version's actual impact on learning outcomes. Education product teams will reassess AI tutor entry design, embedding dialogue in existing practice and course content rather than as a standalone feature. Schools and teachers adopting AI tutoring need to separate overall platform effects from the AI tool's own effects and ask vendors for verifiable evaluation data.