A viral post on August 22 said Harvard Business School had just launched AI “clones” of professors that could hear startup pitches, simulate sales calls, and run mock board meetings. The first part has a real basis. The timing and capability bundle do not.
The verified product is HBS Foundry, an AI-native venture-building workspace. Foundry participants can practice pitches with faculty-modeled AI mentors, including video avatars, then repeat the exercise before meeting a real investor. That is a meaningful experiment in AI-assisted experiential learning. It is not evidence that Harvard has replaced professors, nor that every business-school interaction in the viral claim is available through those avatars.
The distinction matters because this is a better product story than “professor replaced by clone.” HBS is using AI to make practice abundant while keeping high-stakes teaching, mentoring, evaluation, and investment decisions human.
TL;DR: what is real and what was inflated?
| Question | Verified answer |
|---|---|
| Did HBS launch this on August 22, 2026? | No. HBS's fiscal 2025 annual report says Foundry launched during fiscal 2025, and faculty publicly discussed an AI clone in May 2025. |
| Are there AI versions of HBS professors? | Yes. HBS describes personalized AI mentorship modeled on faculty, while Foundry has shown video avatars used for founder pitch practice. |
| Can founders pitch to them? | Yes. Foundry says the avatars ask follow-up questions and provide advice grounded in faculty expertise and HBS frameworks. |
| Do the official sources verify sales calls and mock board meetings? | No. The HBS material reviewed verifies pitches, startup-building support, simulations, and live AI exercises, but not that full capability list for professor avatars. |
| Are the avatars replacing professors? | No replacement is announced. Foundry pairs AI mentorship with live sessions from real experts, and HBS faculty still lead teaching and discussion. |
| Is this part of normal HBS tuition? | The verified avatar use is tied to Foundry's founder programs, not presented as a substitute for the residential MBA or its faculty. |
What Harvard Business School actually built
Foundry's current site describes an eight-week founder bootcamp that moves participants from an early idea to a pressure-tested pitch. The surrounding workspace supports market validation, business-model development, pitch-deck creation, milestone tracking, and AI mentorship modeled on HBS faculty, plus live sessions with real experts.
The clearest official description of the avatars comes from HBS Foundry's own account. Foundry said participants practice pitching to AI avatars of professors and quoted managing director Elise Bates calling the value “low-stakes reps” before a real investor conversation. A participant described taking roughly 30 practice attempts and receiving follow-up questions from the Jeff Bussgang avatar.
That interaction is more than a prerecorded lecture. It combines:
- a visual and voice interface modeled on a participating faculty member;
- a conversational model that can respond to a founder's specific pitch;
- HBS research, frameworks, and faculty expertise supplied as grounding context;
- follow-up questions that force the founder to defend an assumption; and
- repetition without scheduling 30 human office-hour appointments.
Professor Jeffrey Bussgang had already referenced the system in a May 2025 HBS alumni resource, writing that his students “rave about my AI clone.” HBS's fiscal 2025 annual report says Foundry's alpha reached more than 500 founders across six partner organizations. So the August post is resurfacing and broadening an existing program, not documenting a same-day launch.
The headline conflates three different HBS AI efforts
“Harvard launched professor clones” compresses several distinct systems into one dramatic sentence.
| HBS initiative | What it does | Is it a professor clone? |
|---|---|---|
| Foundry faculty-modeled mentors | Founder guidance and repeatable pitch practice, including video-avatar interactions | Sometimes, in the sense of a disclosed faculty likeness and persona |
| Course and tutor bots | Answer course-specific questions and support preparation using class materials | Usually no; these are purpose-built assistants, not replicas of a professor |
| MBA simulations and live exercises | Put students inside decision scenarios across entrepreneurship, marketing, and organizational behavior | Not necessarily; an AI character or simulation is not automatically a faculty clone |
HBS's 2025 Year in Review describes course bots, tutor bots, and an AI data scientist in the required Data Science and AI for Leaders course. Separately, HBS faculty told The Harvard Crimson that the school uses AI simulations, avatars, and live exercises across the MBA curriculum. Those tools can all support active learning, but calling all of them “professor clones” hides how differently they are designed and governed.
This distinction also helps educators. A course bot needs source fidelity and good refusal behavior. A role-play character needs scenario realism and consistent objectives. A faculty avatar adds separate questions about likeness, voice, disclosure, and whether users mistake generated advice for the professor's personal judgment.
Why pitch practice is the strongest use case
Pitching is a performance skill. Reading another framework does not prepare a founder for the moment an investor interrupts with, “Why now?”, challenges market size, or asks what evidence would disprove the thesis. Improvement requires producing an answer under pressure, receiving feedback, and trying again.
That makes the Foundry avatars closer to a flight simulator than a lecture replacement. They reduce three costs:
- Scheduling cost: an AI counterpart is available when the founder is ready to rehearse.
- Social cost: a weak first attempt does not consume scarce investor or faculty attention.
- Repetition cost: the learner can run the same pitch 20 or 30 times and compare responses.
This is the same design line visible in Dartmouth's AI-graded quiz pilot: the useful model does not perform the learner's task. It creates more opportunities for the learner to perform and get feedback. It is also why unguarded AI homework help can reduce unassisted exam performance. If an avatar rewrites the pitch for the founder, it removes the practice. If it asks the next hard question and waits, it creates the practice.
Practice is not proof that the feedback is correct
The public material establishes availability and intent, not educational effectiveness. HBS has not published a benchmark showing how often a faculty avatar's feedback matches the professor's own feedback, how stable the avatar is across repeated prompts, or whether founders who use it raise money or make better decisions at higher rates.
That does not make the tool useless. It changes what can responsibly be claimed.
A credible evaluation should test at least four things:
| Evaluation question | Useful measurement |
|---|---|
| Does the avatar represent the professor accurately? | Blind agreement study comparing avatar and faculty feedback on the same pitches |
| Does repetition improve the learner? | Rubric-scored first versus final pitch, rated by humans who do not know the condition |
| Does feedback transfer? | Performance with a new human investor, not just a higher in-tool score |
| Does the system fail safely? | Rate of invented facts, overconfident legal/financial advice, and appropriate escalation |
This is where AI education products often outrun their evidence. A realistic face and familiar voice can make uncertain feedback feel more authoritative than plain text. Builders should therefore treat identity fidelity and advice accuracy as separate evaluations. Looking like a professor does not prove the answer is what that professor would say.
Privacy, recording, and consent need more than a reassuring headline
Foundry's privacy statement is unusually direct on several important points: founder data is not sold, licensed, or used for model training; participants retain their intellectual property; and the platform says it uses encryption, role-based access, and confidentiality controls.
Those commitments matter because a pitch may expose an unfiled invention, customer names, unit economics, fundraising plans, or regulated data. “Not used for model training,” however, is only one layer of the privacy stack. Public materials reviewed for this article do not spell out:
- how long pitch audio, video, and transcripts are retained;
- which model and avatar vendors can process or access those recordings;
- whether participants can delete a specific rehearsal and its derived feedback;
- how faculty approve the persona, likeness, voice, and future updates;
- how the product distinguishes an avatar's generated advice from the professor's endorsed views; or
- what happens when a participant asks for legal, medical, investment, or other high-stakes advice.
Those are not reasons to reject the experiment. They are the next questions a serious institution should answer. The AI literacy curriculum explainx.ai recommends starts with exactly this habit: inspect provenance, permissions, and failure modes before trusting a fluent interface.
What remains human, and why tuition is the wrong comparison
The social reaction asked whether tuition would stay the same if an AI can imitate a professor. That assumes the scarce value of business school is access to a faculty voice saying plausible sentences. It is not.
Human faculty still:
- design the scenario and decide what good judgment looks like;
- connect a student's answer to live classroom context;
- notice uncertainty, incentives, group dynamics, and ethical blind spots;
- revise frameworks when reality contradicts them;
- make admissions, grading, mentoring, and institutional decisions; and
- take responsibility for consequential advice.
Foundry's own design supports that reading. The program advertises faculty-modeled AI mentorship plus live sessions with real experts, cohorts of real founders, and eventual pitches to real funders. The avatar increases the number of rehearsals before the scarce human interaction; it does not turn the interaction into a degree or an investment decision.
This is also why social learning still matters in an AI course. An always-available model can challenge a claim, but peers create accountability, disagreement, and social consequences that a simulation cannot fully reproduce. On explainx.ai, Melo's Quiz, Practice, and Explain Back modes use AI to increase deliberate practice, while workshops and cohorts keep humans in the learning loop.
A responsible blueprint for building AI simulations
The most transferable lesson from HBS Foundry is not “clone a famous teacher.” It is to design a bounded practice environment.
- Name the job precisely. “Investor pitch sparring partner” is testable. “Digital professor” is not.
- Make the learner go first. Ask for the pitch, decision, or explanation before offering a rewrite.
- Ground feedback in a disclosed rubric. Show which framework or criterion produced the critique.
- Disclose the simulation continuously. Do not rely on one onboarding notice when a face and voice encourage anthropomorphism.
- Separate generated advice from faculty endorsement. A model response is not a quotation or personal decision by the professor.
- Provide correction and escalation. Let learners flag a bad response and route consequential questions to a human.
- Measure transfer outside the tool. A polished in-app score is meaningless if the learner cannot handle a new human question.
- Minimize sensitive data. Give users deletion controls and explain retention and vendor access in plain language.
That blueprint aligns with the broader evidence behind interactive AI learning pathways: AI works best when it structures action, feedback, and reflection rather than replacing the action itself.
The bottom line
Harvard Business School has a real and interesting AI-avatar program. HBS Foundry lets founders rehearse startup pitches with faculty-modeled AI mentors, receive follow-up questions, and repeat the exercise before facing a human investor. The program predates the August 22 viral post, and official sources do not verify the claim's full sales-call and mock-board-meeting bundle.
The consequential shift is not “professors are now software.” It is that a learner can get 30 low-stakes rehearsals before one high-stakes human conversation. That is valuable if the simulation is grounded, measured, private, and clearly disclosed. Without those controls, the professor's likeness risks adding confidence faster than the system adds accuracy.
Related on explainx.ai
- The research behind AI-graded quizzes: 20 studies on interactive textbooks
- What should schools teach when AI can answer every question?
- Dartmouth's Phosphor study: what a 0.71–1.30 SD effect actually did
- The generative AI learning penalty: homework up, exams down
- Social learning for AI: why dialogue still matters
- Introducing Melo, the explainx.ai learning copilot
- Introducing interactive AI learning pathways
- AI curriculum for college students
Official sources: HBS Foundry overview · HBS Foundry privacy and organization program · HBS fiscal 2025 annual report · HBS 2025 Year in Review · AI Integration at HBS · HBS Foundry's avatar statement
Foundry features, privacy statements, program availability, and public evidence reflect HBS materials available on August 22, 2026. HBS may change the platform, policies, or enrollment terms after publication.