Harvard Business School Puts AI Clones in the Classroom

When Your Pitch Coach Is a Digital Twin

Harvard Business School has begun deploying AI-generated avatars of its faculty to provide individualized coaching inside its entrepreneurship bootcamp. The eight-week HBS Foundry program, priced at $699, pairs weekly live sessions with human instructors alongside synthetic versions built by HeyGen, a startup specializing in digital human avatars. Students rehearse investor pitches and board meetings with these avatars, which offer real-time feedback in the voice and likeness of actual venture capitalists and professors.

The move reflects a pragmatic calculus for elite business education: demand for access far outstrips faculty bandwidth, and generative AI offers a way to simulate one-on-one attention at scale. At DailyTechWire, we've tracked similar experiments across Asia's top universities - NUS in Singapore and IIT Bombay have piloted AI teaching assistants - but Harvard's deployment is notable for putting synthetic humans, rather than text chatbots, at the center of the learning loop.

One test case involved Flybridge Capital co-founder Jeff Bussgang, whose digital replica evaluated a deliberately absurd pitch for "Uber for bananas." The avatar delivered critical feedback, though observers noted a conspicuously static smile during the exchange. Bussgang himself called the experience "a little creepy" but acknowledged that students respond positively to the format.

From Chatbot to Avatar

Project director Katharina Rings initially envisioned a text-based assistant - a Q&A interface students could query for advice on pitch structure or fundraising tactics. After Harvard released a trial version of the bootcamp, however, participants pushed back. They wanted structure, not open-ended conversation. They wanted something that felt like a person across the table, not a search bar.

HeyGen's technology answered that demand. The platform generates photorealistic video avatars that mimic speech patterns, facial expressions, and mannerisms of real individuals. For HBS Foundry, that meant students could rehearse a pitch to a digital version of a named investor, receive verbal critiques, and iterate - without booking calendar time or worrying about first-impression stakes.

The preference for avatars over chatbots suggests a behavioral insight: learners treat text interfaces as tools but treat humanoid interfaces as social actors. That distinction matters for practice scenarios where emotional realism - nervousness, timing, body language - shapes performance. A chatbot can tell you your pitch is too long; an avatar makes you feel the awkwardness of a frozen smile when your story drags.

The Instructor Perspective

Faculty reactions have been mixed. Bussgang, who agreed to be cloned, framed the technology as an extension of his teaching reach. His avatar allows dozens of students to receive personalized feedback simultaneously, a logistical impossibility for a single human. He noted that students "love it," a sentiment echoed in participant interviews.

Still, the uncanny valley remains. Observers have pointed out micro-failures in the avatars' facial animations - frozen expressions, slightly mistimed blinks - that break immersion. These artifacts are less about technical limits than about the inherent strangeness of conversing with a synthetic copy of a real person who exists, is reachable, and might one day watch a recording of your interaction with their digital twin.

The broader question is whether avatar-based coaching can replicate the tacit knowledge transfer that happens in live mentorship: the improvisational follow-up question, the sidebar about portfolio strategy, the moment an instructor senses confusion and shifts gears. HBS Foundry's design acknowledges this gap by preserving weekly live sessions with human instructors. The avatars handle repetition and volume; the humans handle nuance and relationship-building.

Scaling the Ivy League

Harvard's experiment is part of a larger strategy to expand access without diluting brand equity. Traditional HBS programs carry five- or six-figure price tags and require either full-time enrollment or executive leave. Foundry, at $699 for eight weeks, slots into a different market: working professionals, international founders, and early-stage entrepreneurs who want the HBS imprimatur but can't relocate to Boston.

AI avatars make that economics work. A single instructor can be cloned across dozens of concurrent cohorts, providing localized feedback without linear scaling of labor costs. The trade-off is a narrower, more scripted interaction - but for a $699 bootcamp, participants seem willing to accept that boundary.

The model also insulates HBS from a common critique of online education: that it's passive, impersonal, and low-engagement. By embedding avatars that respond to student input in real time, Foundry creates the feeling of interactivity even when no human is in the loop. Whether that feeling translates to learning outcomes remains an open empirical question, but early participant feedback suggests the psychological effect is real.

What Students Actually Want

Student sentiment has surprised some observers. On many campuses, generative AI has been met with skepticism or outright hostility - concerns about plagiarism detection, algorithmic bias, and the erosion of human teaching jobs. Foundry participants, by contrast, have reported enthusiasm for the avatar format.

Part of that enthusiasm may be contextual. Entrepreneurship students are predisposed to experiment with new tools, and a $699 bootcamp self-selects for learners who prioritize flexibility and cost over traditional classroom dynamics. The avatars also serve a low-stakes function - pitch practice rather than graded assessment - which lowers the emotional cost of interacting with a bot.

Another factor is the alternative. Without avatars, students in a remote, asynchronous bootcamp would likely submit recorded pitches and wait days for written feedback. The avatar offers synchronous response and verbal tone, closing the loop faster and preserving some of the social cues that make feedback feel personal. It's not a human instructor, but it's closer than email.

The HeyGen Engine

HeyGen, the startup powering the avatars, has positioned itself at the intersection of generative video and enterprise learning. The company's platform allows organizations to create digital twins of employees, executives, or subject-matter experts by training on video and audio samples. The resulting avatars can deliver scripted content or respond to prompts within defined parameters.

For HBS, HeyGen provided both the avatar generation and the real-time inference layer that lets students have back-and-forth exchanges. The technical challenge is latency: generating naturalistic video responses fast enough to sustain a conversation without awkward pauses. HeyGen's architecture combines pre-rendered avatar templates with on-the-fly lip-sync and expression mapping, a hybrid approach that balances realism and speed.

The business model is subscription-based, with pricing tied to usage volume and customization depth. Harvard's deployment represents a high-profile reference customer, likely accelerating HeyGen's traction in the education vertical. Across Asia, we've seen similar platforms - Synthesia, D-ID, and regional players like China's Surreal - compete for enterprise contracts in training, onboarding, and customer support.

Open Questions

Several uncertainties linger. First, how durable is student enthusiasm once the novelty fades? Early adopters of any new technology tend to report higher satisfaction than later cohorts, and avatar-based learning may follow the same curve.

Second, what happens when something goes wrong? AI systems can generate inappropriate, biased, or nonsensical output. In a pitch-practice scenario, a misleading critique could send a founder down the wrong strategic path. HBS has not publicly detailed the guardrails or oversight mechanisms in place to catch such failures.

Third, what are the intellectual property and consent boundaries? Bussgang agreed to be cloned, but the legal and ethical framework for digital twins remains unsettled. If an avatar trained on an instructor's voice and likeness says something the instructor disagrees with, who is accountable? If a student records and shares their avatar session, does that violate the instructor's rights?

Finally, there's the question of what gets lost. Teaching is not just information transfer; it's mentorship, encouragement, challenge, and the transmission of professional norms. An avatar can simulate feedback, but it cannot model curiosity, vulnerability, or the lived experience of building a company. As business schools lean further into AI-mediated instruction, they will need to articulate clearly what they believe cannot be automated - and ensure those elements remain central.

The Broader Trajectory

Harvard's move is unlikely to remain isolated. Other business schools, coding bootcamps, and executive education providers are exploring similar architectures. The economics are compelling: avatars let institutions serve more students without hiring more faculty, and they let students practice without fear of judgment from a human observer.

At DailyTechWire, we expect to see avatar-based learning spread fastest in skills training - sales coaching, interview prep, language practice - where repetition and low-stakes feedback are valuable. The model is less suited to seminar-style discussion, collaborative problem-solving, or any domain where the goal is to surface ideas no one has articulated yet.

The deeper shift is cultural. A generation of learners is now encountering AI not as a curiosity or a threat, but as a normal part of the educational infrastructure. They will graduate expecting digital twins, real-time synthesis, and on-demand expertise. The institutions that thrive will be those that figure out how to use these tools to amplify human teaching, rather than replace it - and that remain honest about the difference.