Who bought the school?
When the companies selling AI tools also build the credentials that certify them, it isn't disruption — it's enclosure.
In April, Khan Academy founder Sal Khan conceded that the AI tutoring revolution he forecast in 2023 hadn't materialised. Khanmigo, the chatbot Khan Academy built on early access to GPT-4, was — in his own description — a "non-event" for most students. They didn't use it. Teachers who once championed it drifted away. Kristen DiCerbo, Khan Academy's chief learning officer, was blunter still: she isn't seeing the revolution in education.
The usual response to a product nobody chooses is to ask why and rebuild it. What happened next was different. Within days, Khan Academy, TED and ETS announced the Khan TED Institute — a new sub-$10,000 degree designed for the AI economy, with applications due to open in 2027. Its corporate "thought partners," tasked with shaping the competency signals the credential will carry, include Google, Microsoft, Accenture, Bain, McKinsey and Replit. No university is involved.
Strip away the announcement language and a structural problem comes into focus. The companies that sell AI tools are now helping build the institutions that certify what those tools produce. It's the logic of a pharmaceutical company running its own clinical trials and submitting the results as independent proof. A more accurate word than disruption is enclosure: the shared, slowly built understanding of what a degree certifies is being moved onto private ground.
None of this means the problem the Institute points at is invented. The gap between what employers want and what graduates leave with is real and decades old. But scale is not legitimacy. South Korea spent roughly $385 million on AI textbooks; adoption stalled near 30% and lawmakers stripped the programme's legal status within months. And the evidence on AI as a learning tool is not flattering. A University of Pennsylvania study of nearly 1,000 Turkish students found that those given a standard chatbot to practise with solved far more problems during the session — then scored 17% worse than peers who never used AI once the tool was taken away.
That points to a quieter cost. Entry-level software jobs for 22-to-25-year-olds have fallen close to 20% since ChatGPT launched, on Stanford's AI Index figures. Those are the roles where judgment is built by doing the work badly before doing it well. A credential redesigned by companies whose products remove that struggle optimises for the certificate and hollows out what it was meant to mean.
Better models exist. The Australian Centre for Student Equity and Success built its AI framework slowly and on evidence; ETH Zurich and EPFL released a fully open 70-billion-parameter model on public infrastructure; Stanford Medicine built its own clinical-reasoning tool rather than renting one. Each kept the institution in control of the goal. Which leaves one question worth holding onto: we wouldn't let a student grade their own exam, so why let the vendor issue the degree?
Happy to add a one-line "full conversation on Adjunct Intelligence" tag at the end if you want it pointing back to the episode — left it off so the piece stands on its own.