On March 14, 2023, the day after OpenAI released GPT-4, the nonprofit Khan Academy announced Khanmigo, an AI assistant built on GPT-4 and offered in a limited pilot to teachers, students, and donors. Sal Khan positioned it in two roles: a one-to-one tutor for students and a teaching assistant for educators that could help draft lesson plans and assignments.
The product’s design choice was deliberate and pedagogical: rather than simply answering a student’s question, Khanmigo is prompted to behave like a tutor, asking the student questions and working through a problem in dialogue. Khan emphasized the organization’s nonprofit framing, arguing that its “North Star is driving more learning, not driving shareholder value or profits.” The announcement was candid about limitations, noting that the model can still “hallucinate” - make things up - and that Khan Academy added its own guardrails to reduce the math errors GPT-4 is prone to.
The launch was widely read as the moment generative AI entered mainstream K-12 education, and it explicitly invoked the ideal of personalized tutoring that has animated education technology since Bloom’s 2-sigma paper.
Why business readers should care: Khanmigo is the canonical example of wrapping a general-purpose LLM in domain-specific prompting and guardrails to make it fit for a regulated, high-trust use. The value was not the raw model - anyone could call GPT-4 - but the pedagogy and safety layer built on top.