Khan Academy Personalizes Education with AI Tutor. How it Works

Which company's case study is this?

Khan Academy is a US non-profit organization with the mission 'to provide free, world-class education for anyone, anywhere.' Based on founder Sal Khan's belief that 'education is a fundamental human right,' it offers high-quality learning content worldwide. In recent years, it has focused on developing 'Khanmigo,' an AI tutor powered by generative AI, aiming to provide one-on-one instruction to all learners and bridge educational disparities.

Challenges to be addressed

In the educational field, while the importance of personalized instruction tailored to each student's learning pace and comprehension is recognized, its implementation has been difficult given the reality of one teacher being responsible for dozens of students. Furthermore, high-quality instruction, such as the 'Socratic method' which encourages students to think and arrive at answers themselves rather than simply being given the solution when they get stuck, requires skilled techniques and time. Moreover, teachers spent a lot of time on administrative tasks such as lesson preparation, assignment creation, and evaluation, making it challenging to secure time for direct interaction with students.

How AI was used

Khan Academy developed the AI tutor 'Khanmigo' on OpenAI's GPT-4, as one of OpenAI's earliest GPT-4 launch partners. The system has since evolved into a multi-model architecture: the core tutoring engine continues to run on GPT-4-class models, while a partnership with Google announced in January 2026 brought Gemini to the 'Writing Coach' feature, and a partnership with Microsoft provides Khanmigo for Teachers free to U.S. teachers and explores the small model Phi-3 for lighter tasks. For students, Khanmigo functions as a tutor that doesn't directly provide answers but encourages thinking by offering hints through dialogue. It supports a wide range of subjects, including mathematical problem-solving, essay writing coaching, and coding debugging assistance. For teachers, it acts as an assistant that supports administrative tasks such as lesson plan creation, automatic assignment generation, and student progress summary creation, thereby reducing their workload.

Proposed Architecture

The following is an estimated architecture derived from publicly available information and common configurations. The division of foundation models (GPT-4, Gemini, Phi-3) is confirmed by official announcements, but details of the internal system integration remain inferred and are not definitive.

• ▸Input Data: Student chat questions, essays written by students, teacher instructions for lesson plan creation, etc.
• ▸AI Processing/Model or Retrieval Layer: The core conversational engine runs on OpenAI's GPT-4-class models (presumably via Azure OpenAI, given the Microsoft partnership). A multi-model configuration has been publicly announced: Google's Gemini powers the 'Writing Coach' feature, and the small model Phi-3 is used for lighter tasks. It is presumed that a mechanism like RAG (Retrieval-Augmented Generation) is used to generate contextually relevant educational responses by linking with Khan Academy's vast educational content (articles, videos, practice problems). Safety filters are also integrated to prevent inappropriate responses.
• ▸Output to Business Systems: Hints and questions for students, lesson plan proposals for teachers, etc., are presented via Khanmigo's chat interface. Some functions are also integrated with Learning Management Systems (LMS) like Canvas.
• ▸Monitoring and Governance: Teachers can select the AI usage mode (interactive mode or feedback only) and control the degree of AI intervention. Additionally, conversations between students and the AI can be reviewed by teachers, which can be useful for educational guidance. To eliminate bias, the AI model is continuously monitored and updated.

Implementation Effects and Key Takeaways

• ▸Observable Effects and Changes: It has been reported that the time teachers spend on lesson preparation has been reduced, leading to an increase in direct interaction time with students. In New Hampshire, USA, a statewide Khanmigo pilot program has been implemented, with approximately 5,000 educators and 40,000 students using it, indicating widespread adoption.
• ▸Key Points for Readers: The AI is designed not merely to answer questions with knowledge but as a 'coach' that emphasizes the learner's thought process and fosters the ability to think for oneself. It's also crucial that AI is positioned not as a replacement for teachers, but as an 'assistant' to support them and enable more human-centric education.
• ▸Points of Caution: Continuous attention is required for ethical issues such as ensuring the accuracy of AI-generated information, handling personal information in educational settings, and addressing the digital divide.

What Japanese Companies Can Learn

The Khan Academy case demonstrates that combining a company's specialized knowledge and content with generative AI can offer new value to customers. The concept of 'coaching AI,' which encourages customer growth and self-resolution rather than being a mere Q&A system, is an idea applicable not only in education but also in many business areas requiring specialized knowledge, such as customer support, employee training, and knowledge transfer.

Today's AI News

Reference Links

Find Related AI Adoption Cases