Harvard's $699 Startup Bootcamp Offers AI Avatars of Its Instructors

Harvard Business School launched a new startup bootcamp in August 2026 called HBS Foundry — priced at $699 for eight weeks — with a feature that distinguishes it from the crowded online education market: AI-generated avatars of its real instructors, built to give students personalized feedback on their pitches. The program isn't a vanity experiment. Multiple reports confirm it's already running, with paying participants, and the results are surprising enough to warrant close attention from anyone building or evaluating AI-augmented learning tools.

What Changed

For years, the promise of AI in education has centered on scalability — the idea that a single great teacher could reach thousands of students without degrading quality. HBS Foundry represents one of the first institutional attempts to test that premise at a credible price point, using avatar technology rather than static video or chatbot interfaces. The bootcamp runs eight weeks, costs $699, and is aimed at entrepreneurs, not enrolled degree students. That positioning matters: it gives Harvard a lower-stakes environment to experiment with formats that would be politically difficult to introduce into its core MBA curriculum.

The avatars themselves were created by HeyGen, an AI video startup that has built a reputation for producing photorealistic digital doubles. HeyGen's involvement is confirmed by multiple sources, and its role here is infrastructure, not curriculum — HBS brought in an existing vendor rather than attempting to build its own avatar technology. That vendor relationship is a meaningful signal: elite institutions are now comfortable enough with third-party AI video platforms to attach their instructors' likenesses and reputations to them.

The bootcamp's structure draws a deliberate line between AI and human responsibility. According to reports, students have live sessions with real instructors every week. The AI avatars handle the practice rounds — pitch rehearsals and simulated board meetings — where repetition and iteration matter more than the relationship. Project director Katharina Rings is credited across sources with designing this division, though the final product looks different from what she originally intended. She initially conceived of the AI component as something closer to a chatbot. The avatar format emerged from student feedback during a trial run, not from the original product plan.

Notably, the $699 price sits at an unusual intersection: too expensive to be dismissed as a commodity course, cheap enough to be accessible to early-stage founders who couldn't afford HBS programming any other way. Whether that price holds as the program scales, or whether it subsidizes a proof-of-concept HBS plans to productize further, is not yet clear from any of the sources reviewed.

How It Works

The clearest window into the HBS Foundry experience comes from a session documented by New York Times reporter Sarah Kessler, who pitched an AI-generated copy of Jeff Bussgang, co-founder of Flybridge Capital and one of the instructors whose likeness was built into the system. Kessler's test pitch — a startup she described as "Uber for bananas" — didn't impress either version of Bussgang. The real instructor and his avatar apparently reached the same verdict. The detail worth noting: Kessler observed that the virtual Bussgang maintained a noticeably frozen smile during her pitch, a visible artifact of the avatar technology. It's the kind of thing that registers as uncanny in an isolated moment but apparently doesn't undermine the overall experience for enrolled students.

That gap between a journalist's one-off trial and a student's extended engagement with the system is worth taking seriously. Kessler's experience was adversarial by design — she was testing the system. Students using the bootcamp for real stake feedback over eight weeks are in a different psychological position. They've paid money, they have goals, and the avatar is a tool for them to improve, not a subject for skepticism. Context shapes perception here in ways that make controlled user research hard to replicate from the outside.

The product evolution is also instructive. Rings and the HBS team ran a trial before committing to the avatar format, and the feedback they got was unexpected: students wanted the AI to be more guided, not more open-ended. That preference pushed the team away from a conversational chatbot design toward the structured avatar format — something that feels more like a real interaction, with a face, a presence, and a defined role. Our read is that this finding has implications well beyond this single program. It suggests that users approaching AI for skill development may tolerate the uncanny valley more readily than they resist the absence of structure. A formless chatbot can feel like work. A face that responds, even imperfectly, can feel like practice.

What It Means for Developers

For developers and teams building educational AI tools, HBS Foundry is a useful reference point for what "AI-augmented instruction" can look like at institutional scale. The architecture here isn't novel — avatar technology, vendor APIs, structured curriculum — but the combination is running in production at a named institution with real students and a real price tag. That matters for credibility benchmarking. If you're pitching an AI tutoring product to an enterprise or university buyer, the HBS Foundry example is now a comparable you can reference without inventing one.

The HeyGen relationship is worth watching separately. The company has positioned itself primarily as a marketing and video localization tool. A deployment inside a Harvard bootcamp — where the avatar is giving substantive feedback on startup pitches, not narrating a product demo — is a meaningful expansion of the use case. If HeyGen can sustain quality in that kind of interactive, high-stakes context, the platform's addressable market in corporate training and university programming becomes considerably larger. A separate source notes that the open question is whether the HBS model extends beyond the $699 bootcamp into formal degree programs; if it does, the avatar vendor sitting at the center of that infrastructure has a durable foothold.

For teams specifically building practice-and-feedback loops — interview prep, sales coaching, pitch training — the student preference data is the most actionable signal in this story. Users wanted more guidance, not less. They chose the face over the chat window. That preference may not generalize to every domain or user population, but it's a real data point from a real trial, not a design assumption.

Bussgang, for his part, acknowledged the strangeness of the situation plainly. He called his digital copy a little "creepy" — and then confirmed his students love it. That tension, between the creator's discomfort and the end user's utility, is probably the most honest summary of where AI avatars in education currently stand.

Sources