Expert takes · This week (published Jul 23)
From the Jul 24, 2026 daily brief
Mollick, a professor at Penn's Wharton School and the most widely read AI pragmatist among business audiences, gives a hands-on read for the back half of 2026: for real work, the practical choices are ChatGPT or Claude (from $20/month), and the difference lives in the agent execution environment and product scaffolding, not the models themselves. He sorts each company's products into two types: those that deliver results (the model works in a cloud virtual machine and hands back decks, analyses, and files for your review) and those that show their work (they take over your computer, and you watch them edit files and run commands). His verdict on Google is harsh: it "has fallen behind where it now counts: it has no leading frontier model and it has nothing close to Codex and Code" (Codex is OpenAI's process-visible agent); Chinese open-source models are "surprisingly capable" but take real expertise to use as agents (One Useful Thing, Jul 23). This supports our July 19 call that the valuable position in AI is moving away from the model itself: the model layer went multipolar, but usable agent products did not, and the dividing line has moved to the scaffolding layer. Caveat: this is one person's read; Mollick has early-access relationships with the labs, and he says himself the picture can change at any time.
Just an email address, unsubscribe anytime. This is the only thing we ask of you.
In his year-end interview (we only obtained the transcript last night), Google DeepMind CE…
In the same interview he decomposed the AI bubble by layer: "seed rounds for startups that…
NYU professor emeritus and prominent AI skeptic Gary Marcus argues that Chinese model capa…
He went through six open-weight models' technical reports side by side and found that the …