• Take 1: AI can make high-level coaching endlessly repeatable, but it cannot share the personal stakes of mentorship. • Take 2: Foundry’s accessible price widens Harvard’s reach while raising uncomfortable questions about what students are really buying.

Harvard Business School is turning professors into tireless digital coaches. The experiment promises more access to expertise—but also tests how much of elite teaching can be reduced to software.

Universities were already moving in this direction. Georgia Tech created an AI avatar for an online course, while its “Jill Watson” teaching assistant served large virtual classes; Imperial Business School has also built digital twins of faculty.¹

At Harvard, Foundry director Katharina Rings initially imagined something closer to a chatbot. After testing an early version, however, participants asked for a more guided experience.²

The result is an eight-week, $699 startup bootcamp featuring AI versions of seven professors and senior lecturers. Founders can repeatedly rehearse investor pitches, sales calls and board meetings before attending weekly live sessions. Some 760 founders have participated, and the program culminates in a chance to pitch for $100,000.¹

Harvard says the avatars were built through voluntary interviews, recordings, testing and continuing faculty feedback. The aim, Rings said, is to let founders practice before meeting mentors and investors, making those human conversations “even more meaningful.” Professor Shikhar Ghosh was equally explicit: “The goal isn’t to create a substitute for me.”¹

Participants appear receptive. Lecturer Jeff Bussgang conceded that his digital copy is “a little ‘creepy,’” but added, “My students love it.” A reporter testing the avatar found its frozen smile less convincing when pitching an “Uber for bananas.”²

Critics see a hard limit. Education consultant Lindsay Tanne Howe said AI may reproduce a professor’s perspective at scale, “but what it can’t replicate is everything that happens between people.” Stanford lecturer Kian Katanforoosh argued that software cannot truly know a student, risk its reputation on an introduction or “genuinely care what happens next.”¹

That leaves Harvard’s wager unresolved: AI may improve preparation for human teaching, but its efficiency also exposes precisely what the machine cannot provide.