Harvard's $699 Startup Bootcamp Now Uses AI Video Avatars of Its Instructors

Harvard Business School's HBS Foundry, an eight-week startup bootcamp priced at $699, has integrated AI-generated video avatars of its instructors into the program's curriculum. The avatars, built by startup HeyGen, give participants feedback during practice pitches and board meeting simulations. Live weekly instructor sessions continue to run alongside the AI components TechCrunch.

The avatar system did not start where it ended up. HBS Foundry project director Katharina Rings said she initially envisioned the AI component as closer to a chatbot — a text-based conversational tool. After HBS released a trial version, students reported wanting a more guided experience, which pushed the design toward the video-avatar format now in use TechCrunch.

New York Times reporter Sarah Kessler tested the system by pitching to an AI-generated copy of Flybridge Capital co-founder Jeff Bussgang, one of the Foundry's instructors. During the pitch, the virtual Bussgang displayed a noticeably frozen smile TechCrunch. That uncanny-valley effect — the uneasy feeling you get from a face that looks almost human but moves slightly wrong — did not appear to dampen enthusiasm among actual participants. Foundry participants told Kessler they like the avatars TechCrunch.

Bussgang himself acknowledged the creepiness factor. His AI digital copy is "a little creepy," he said, but added: "My students love it." TechCrunch. The tension between that self-aware discomfort and reported student satisfaction sits at the center of what HBS is doing here.

HeyGen's avatars are not replacing the live weekly sessions. They function as a rehearsal layer, giving participants a way to practice pitches and board-meeting scenarios with on-demand feedback outside of scheduled class time. That is a useful application of synthetic video, and it maps onto a pattern educators have explored with text-based AI tutors: the AI handles repetitive, high-frequency interactions while human instructors focus on sessions where their judgment and presence add the most value.

The frozen smile Kessler encountered, though, is a reminder that the technology is not yet transparent. A video avatar that locks into a rigid expression during a simulated pitch creates a different cognitive load than a text chatbot that simply lacks a face. The brain expects facial movement to track conversational flow, and when it does not, the simulation reads as wrong. Bussgang calling his own avatar "creepy" is an honest signal that even the people whose likenesses are being deployed recognize the gap between the synthetic and the real.

Rings's account of the design evolution from chatbot to guided avatar experience is also telling. Students did not want a free-form conversational agent. They wanted structure, direction, and something that felt more like an instructor guiding them through a scenario than an open-ended chat. That preference for guided interaction over unconstrained dialogue is consistent with what has been observed in other AI-in-education deployments, where open-ended access to large language models without scaffolding tends to produce lower engagement and weaker learning outcomes.

The $699 price point for an eight-week Harvard-branded program is notably accessible relative to traditional executive education offerings. The avatar component presumably helps keep costs down by reducing the instructor hours needed per student for repetitive practice scenarios. Whether that economics calculation holds as the technology matures and expectations rise is an open question.

The broader context here is that this is an early data point in what will likely be a widespread pattern: elite institutions using synthetic media to extend the reach of their branded instruction at lower cost and higher scale. The combination of a recognizable faculty member's likeness, a structured curriculum, and on-demand availability is a compelling value proposition for learners who cannot access the in-person version. The quality gap between the avatar and the human, as Bussgang's own discomfort indicates, has not closed. But the students do not seem to mind, and that acceptance matters more than the technical imperfections. When the people actually using a technology report satisfaction despite its visible flaws, the technology has crossed a threshold that no benchmark score can measure.