MasterClass Deploys Multi-Agent AI Tutors to Scale Personalized Learning
MasterClass is building AI teaching agents that adapt to individual student needs, aiming to make one-on-one instruction more affordable and accessible. The effort is driving CoreWeave's production AI infrastructure as companies move beyond experimental deployments.
Bringing personalized tutoring to scale requires more than inserting a chatbot into a classroom. MasterClass is developing AI teaching agents that monitor learner behavior—detecting cognitive overload and waning engagement—and adjust instruction accordingly. The approach addresses a persistent problem: one-to-one tutoring remains expensive, staffing is limited, and quality varies widely.
Mandar Bapaye, chief product officer of Yanka Industries Inc., d/b/a MasterClass, emphasized the importance of grounding AI agents in educational science. "It's not just you throw an agent, throw a chatbot in with a student and let them figure it out," Bapaye said. "Having a very … scientifically and pedagogically backed backbone of your agentic system is of prime importance."
The work is accelerating CoreWeave Inc.'s push into full-stack AI cloud infrastructure as organizations move AI agents from research into live production. Lukas Biewald, senior vice president of AI initiatives at CoreWeave, has advised MasterClass on product development and stressed the need for continuous refinement. "I really think it's through that rigorous evaluation and those loops like we talked about last time," Biewald said. "Let's try a new model, let's try a new rubric, let's see how it does, and let's just keep doing it again and again and again and make it iteratively better every day."
From evaluation to production at scale
MasterClass Executive, the company's new AI-native business program, runs approximately 10 agents in parallel for each learner interaction. The system tracks inputs, outputs, tool calls and agent-to-agent communication. Moving from controlled testing to thousands of simultaneous users introduces new complexity.
MasterClass recently selected W&B Weave to trace, monitor and improve its teaching agents in production. The company built its own agent using Weave's Model Context Protocol interface, which reviews traces each night and surfaces issues with probable causes. "Once things are in production, observability becomes a nightmare because there are thousands of people who are interacting," Bapaye explained. "Now, what you need to figure out is, 'Hey, we did all this good stuff during the eval period, but during production, how are the agents behaving? Are the experiences we are giving really good?'"
Early demand signals suggest the approach is resonating. The first cohort of MasterClass Executive drew 30,000 applications for roughly 500 available spots. The second cohort is approaching 50,000 applications.
https://www.youtube.com/embed/EWBLRiR4eWc?start=5&feature=oembed
Bapaye identified three persistent barriers to personalized instruction that AI addresses simultaneously. "It's cost, it's supply and it's quality," he said. "It's extremely costly to hire personal teachers, they are in very short supply … and the quality of teachers varies. AI solves pretty much all of these three aspects."