Harvard Business School just launched a $699 startup bootcamp where AI avatars of its professors deliver feedback during practice pitches. Around the same time, British lab Inherent released Faraday, an AI agent that outperforms Anthropic and OpenAI at replicating scientific research. Both stories are nominally about AI capability. They are actually about the same anxiety: what happens when expertise becomes a simulation of itself.
Credentials as Content, Professors as IP
The HBS Foundry program is not really selling education. It is selling the aesthetic of HBS mentorship at a fraction of the price, which means it is selling the brand. The avatar does not sweat your bad pitch. It does not have a bad day. It will not remember you at a reunion. What gets lost in the compression is exactly what Daron Acemoglu argued in his 2026 Nature essay: that the AGI framing obscures a simpler and more damaging dynamic, which is that AI is being deployed to replace human judgment in contexts where human judgment is the whole point. A professor's avatar is not a professor. It is a professor-shaped response generator.
Replication as the New Credential
Inherent's Faraday benchmark, meanwhile, measures whether an AI can reproduce the results of published scientific papers. This is genuinely useful, and also quietly alarming: a significant chunk of published science cannot be replicated even by human researchers. A 2026 Nature investigation found dozens of studies invalidated by a single wrong antibody. If Faraday can replicate papers faster than humans, it may also replicate their errors faster. The question of who checks the checker is not answered by a benchmark. Soleio's observation that speed is a moat applies here with some irony: the fastest replicator wins, until it doesn't.