Harvard Business School is adding AI-generated versions of instructors and startup advisers to its eight-week HBS Foundry program, giving participants a way to rehearse pitches and board meetings between live sessions. The bootcamp costs $699 and uses avatars created by HeyGen to provide feedback in simulated startup scenarios.

The move gives Harvard a way to extend the reach of its instructors without turning every interaction into a live class. It also puts a recognizable business school directly into an emerging debate over whether AI representations of experts can offer useful coaching—or merely imitate the appearance of judgment.

How HBS Foundry is using the avatars

According to TechCrunch, HBS Foundry still includes weekly live sessions with instructors. The AI avatars are used separately for practice pitches and board meetings, where participants can present ideas and receive responses from digital versions of program contributors.

The system was initially imagined as something closer to a chatbot, project director Katharina Rings told the publication. After Harvard released a trial version, however, students reportedly asked for a more structured and guided experience. The avatar format became the response to that feedback.

That distinction matters for product teams evaluating AI coaching tools. A chatbot can answer questions on demand, while an avatar-based simulation attempts to create a more specific social setting: an entrepreneur presenting to an investor, or an executive explaining a decision to a board. The value depends less on the avatar’s visual realism than on whether the underlying system can produce relevant, consistent and actionable criticism.

A realistic test also exposed the limits

New York Times reporter Sarah Kessler tested the concept by pitching an AI-generated version of Flybridge Capital co-founder Jeff Bussgang. The pitch, for a business described as “Uber for bananas,” reportedly failed to impress either the real Bussgang or his digital counterpart.

Kessler also noted that the virtual Bussgang displayed a noticeably frozen smile during the presentation. The detail highlights a current limitation of AI avatars: visual or vocal realism can break down at precisely the moments when users are judging whether the interaction feels credible.

Bussgang acknowledged that his digital copy was “a little creepy,” while also saying that his students like it, according to TechCrunch’s account. Participants in the program similarly told Kessler they viewed the avatars positively. Those reactions provide useful context, but they are anecdotal rather than evidence of broad demand or improved startup outcomes.

What the evidence does—and does not—show

The confirmed details in the available reporting are limited but concrete: HBS Foundry is an eight-week program priced at $699; it includes weekly live instructor sessions; and its AI avatars, built with HeyGen, are used for feedback during simulated pitches and board meetings.

The reporting does not establish that the avatars improve fundraising results, business survival, pitch quality or learning outcomes. No independent benchmark, participant completion data or controlled comparison with conventional coaching was provided in the source material. Claims about participant enthusiasm are based on interviews and reported reactions, not a formal adoption study.

That makes the program better understood as an applied experiment in AI feedback than as proof that digital instructors can replace human advisers. The live sessions remain part of the curriculum, and the avatars appear to supplement rather than eliminate human instruction.

Why the model matters to builders and enterprises

For startup founders, the clearest benefit is availability. A participant can practice repeatedly without waiting for an instructor’s office hours or scheduling a new group session. That could make feedback more accessible for routine rehearsal, especially when the goal is to improve structure, clarity or responses to predictable questions.

For enterprises, the same pattern could apply to sales training, management simulations and compliance exercises. A digital representation of a subject-matter expert might help standardize practice across a large workforce. It could also preserve a particular teaching style or institutional perspective after the expert is no longer available for every session.

But deployment raises practical questions that the HBS Foundry example does not answer. Organizations would need to decide how an expert authorizes the avatar, what source material governs its answers, how mistakes are corrected and whether users understand that they are interacting with a simulation. Sensitive business plans create additional concerns around data retention, access controls and model-provider policies.

Reliability is another constraint. A convincing face does not guarantee good advice. If the avatar invents an investor preference, misreads a financial claim or delivers generic criticism, its authority could make the error more damaging. Builders should therefore treat AI feedback as a layer in a workflow with human review, not as an autonomous substitute for experienced judgment.

What to watch next

The most important follow-up signal will be whether HBS Foundry publishes evidence that the avatars improve participant performance. Useful measures could include changes in pitch quality, user retention, repeat practice and differences between avatar-only and instructor-supported cohorts.

Product teams should also watch for clearer information about the technical arrangement between Harvard and HeyGen, including how an instructor’s likeness and voice are licensed, whether instructors can review or correct generated responses, and what happens to participant recordings and business ideas.

A further test will be whether users continue to value the system after the novelty fades. If participants use the avatars repeatedly for specific, measurable tasks, the model may become a practical coaching tool. If the experience is remembered mainly for its frozen expressions or uncanny presence, it may remain a demonstration of generative media rather than a durable education product.

Creati.ai perspective

Harvard’s experiment is notable because it places AI avatars inside a paid, structured learning product rather than presenting them as a standalone entertainment feature. The business case is straightforward: one digital instructor can potentially handle many low-stakes practice interactions while human teachers focus on deeper feedback.

The harder question is whether the avatar adds instructional value beyond a well-designed conversational system. For AI builders and enterprise buyers, the answer should be tested through outcomes, safeguards and repeat usage—not through visual realism alone. HBS Foundry offers an early example of that test, but the available evidence is not yet enough to show that AI avatars can replicate the judgment users seek from real advisers.