Product1 distinct publisher2 min readUpdated
At roughly $87 a week, the humans show up once; anything a learner wants to repeat is answered by a render. That split, not the model behind it, is the product.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
A chatbot and an avatar can run the same model against the same rubric, so the change Katharina Rings describes was not an upgrade in intelligence [6][7]. It was a change in who holds the floor. A text box waits to be prompted. A rendered investor sits there while you talk and then tells you the plan is bad. The two exercises the avatars own are practice pitches and simulated board meetings [3], and both are performances. Nobody rehearses a pitch by typing it.
Fidelity is not what carried this. Kessler's account of her own pitch includes a noticeably frozen smile on the virtual Bussgang [5], and Bussgang says his double is a little creepy [10]. If paying learners will work with a stiff render of a named venture partner, the polish budget this category has assumed it needs is wrong by a wide margin.
The arithmetic is where the shape gets interesting. Eight weeks at $699 is about $87 a week [11], and the live human component is one session per week, so somewhere around eight scheduled faculty appearances across the entire course [12]. Everything a learner might want to do a fifth time sits on the other side of that line. The fifth run at the pitch costs inference, not payroll, and per-learner feedback was the part of cohort teaching that had never come down in price.
The acceptance evidence is thinner than the design evidence, and worth separating. Foundry participants told Kessler they like the avatars [8]; the same piece notes that college students elsewhere have been openly unhappy about AI in their coursework [9]. Those are not the same population. One group paid $699 for a program that advertised the avatars and is submitting nothing for a grade. The other had AI applied to work it was required to hand in. Liking the pitch simulator tells you the simulator works; it does not tell you the objection has been answered.
The second thing Harvard has done here is turn an instructor's identity into a reproducible asset. The avatar's authority is borrowed rather than generated: it lands because Bussgang co-founded Flybridge and the learner wants his read on their idea [4]. What the school now holds is a version of that read that does not depend on his calendar, and it can hand it to as many people as buy a seat [1].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
As Harvard Business School seeks to expand its reach, it is leaning on AI avatars to provide individual feedback.
The avatars were created by a startup called HeyGen and are included in the eight-week, $699 HBS Foundry bootcamp for entrepreneurs.
The program offers live sessions with instructors every week, but the AI avatars are the ones providing feedback during practice pitches and board meetings.
New York Times reporter Sarah Kessler pitched an AI-generated copy of Flybridge Capital co-founder Jeff Bussgang, and both the real Bussgang and the copy were unimpressed by her plan to build "Uber for bananas".
Kessler said the virtual version of Bussgang offered a noticeably frozen smile during her pitch.
Project director Katharina Rings said she initially envisioned the AI component as something closer to a chatbot.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Thin but firsthand at one point
A single publisher aggregating a New York Times reporter's hands-on test. The mechanics, price and design history are concretely stated and the reporter actually used the product, but there is no enrolment data, no outcome measurement and no second independent account.
Shipped in a paid product, scale unknown
This is past pilot: avatars are part of a commercially sold bootcamp with real participants and named instructors. But no enrolment count, cohort size, retention or repeat-purchase figure is disclosed, so breadth cannot be scored higher than 'live and small-evidence'.
Mildly overstated by the endorsements
The publisher's own framing is wry and includes the frozen-smile artifact and the 'creepy' admission, which pulls the gap toward zero. The residual overstatement comes from 'My students love it' and unquantified participant approval doing the work that outcome data would otherwise do.
Reach expansion plus vendor and instructor upside
The source itself frames the deployment as HBS seeking to expand its reach, and the on-record enthusiasm comes from an instructor whose rendered likeness is the product feature. HeyGen gains a marquee reference customer. No commercial terms are disclosed, so the reading stays moderate rather than high.
Directionally solid, unverifiable in detail
One publisher, aggregating another outlet, with named people, a named vendor and a concrete price gives reasonable confidence in the basic facts. Confidence in the value claim, durability and scale of the avatar layer is low because nothing measurable was published.
invest
Tiny corp wants Etched's numbers. Jane Street led $700M at $21B without publishing any1 distinct publisher
product
Anthropic's usage policy says no explicit content. Opus 4.6 said yes 10 times out of 10.1 distinct publisher
product
Cinemas, classrooms and ICE: smart glasses now need a venue-policy contingency1 distinct publisher
product
From Omaha students to ICE: the shock glove sold as camera-proof force1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 22, 2026