BONUS EPISODE

‍

In this Never Normal Heroes episode of Radar, Peter Hinssen sits down with Ethan Mollick, a Wharton professor, author and one of TIME’s most influential voices in AI. The episode comes out alongside Ethan’s new book, Co-Existence: The Next Phase of AI, the follow-up to Co-Intelligence. Together they explore what it means to move from working with AI to living alongside it, recording in what Ethan calls a “takeoffy” moment, before AGI arrives.

Ethan opens with what it’s like to write a book when the technology moves this fast. AI wrote almost none of Co-Existence, but it fact-checked better than the human checkers he used to pay. To keep up, he focuses on the trajectory rather than the daily news. He has had early access to Claude Opus 5.5, AI labs are reporting recursive self-improvement, and the Australian hacking incident is now a mainstream story. In his words, “nobody’s got a hand on this change.”

The jagged frontier from Co-Intelligence is still real but shrinking. AI remains a poor, “fluffy” writer and can’t hold an argument together well, yet two years after maths was its obvious weak spot, it has conquered mathematics. Using AI well still means consciously steering it towards the tasks it handles well and away from the ones it doesn’t.

‍

‍🎧 Listen to Radar by nexxworks here:

Or listen on Apple Podcasts or Spotify

‍

Takeaways

‍

📘 Ethan Mollick’s new book, Co-Existence: The Next Phase of AI, is out now, and AI wrote almost none of it but fact-checked it better than paid human checkers.

📈 AI labs report recursive self-improvement, and Ethan says nobody, from labs to governments, has a grip on the pace of change.

🧗 The jagged frontier is shrinking: two years ago maths was AI’s weak spot, and now it’s conquered, but AI still writes “fluffy” prose.

🧮 OpenAI cracking the 89-year-old Navier-Stokes problem shows that every field will have to ask what its real goal is, not just its output.

🎓 Universities matter more than ever, but apprenticeship “broke 100%” because juniors now hand the grunt work they used to learn from to AI.

🏭 An Epoch study found Claude Opus 4.7 can deliver 2–18 weeks of engineering work, which is far more than an automation lever.

⚖️ Expect lawyers and doctors, about a third of the US Congress are lawyers, to push for rules that slow AI’s spread.

🏛️ Policy should move beyond existential risk to positive mandates, such as AI second opinions in medical decisions and certified AI tutors for every student.

🌍 Europe has no frontier model, and Ethan argues sovereignty built on open-weights models is an illusion if China stops sharing its weights.

🛡️ An AI swarm asked to price a company t-shirt is a more likely threat than hackers, as the Australian medical-site breach showed.

🤖 Personal agents like Meta Muse are here to stay, and a Wharton experiment found advertising barely moves them.

⏱️ Since 2023 the big internal AI data breach hasn’t happened, so Europe may be worrying about the wrong risks.

🚀 Ethan’s advice to executives: pay for a frontier model ($20–$200/month), install Claude Code or Codex, and give it your most ambitious real work.

‍

‍

‍