Mollick argues new AI models approach "jagged AGI" across coding, business, and humanities
Fable 5.2 demonstrates expert-level performance in narrow domains but still shows gaps in judgment and taste, according to the author's hands-on evaluation.
1 source · cross-referenced
- A new AI model, Fable 5.2, performed at or above expert human levels in coding, business planning, and humanities course creation, per a first-person evaluation.
- The author introduces the term "jagged AGI" to describe near-expert performance across many computer-based intellectual tasks while acknowledging remaining gaps.
- Evaluations included building hardware hacks, drafting a plausible business plan, and reconstructing a notoriously difficult humanities syllabus.
Ethan Mollick of One Useful Thing evaluates Anthropic’s Fable 5.2 model ahead of its public release, describing it as "very impressive" and building on the earlier Opus 5.5. He proposes a working definition of "jagged AGI" as "an AI as good as most human experts at most intellectual tasks you can do with a computer," explicitly excluding physical-world interaction and many real-world jobs.
Mollick tests Fable 5.2 across coding, business planning, and humanities scholarship. In coding, he asks the model to transform a USB keyboard into a game console using undocumented hardware features; Fable 5.2 reverse-engineers the device’s API, writes six playable games, and delivers a working prototype from a single prompt. In business, he tasks the model with launching a revenue-generating venture requiring no more than 20 minutes of his time per month; Fable 5.2 screens 22 ideas, selects "Prebid Brief," a service tracking municipal funding for equipment sales reps, and drafts a full marketing plan, pricing, and infrastructure outline. In humanities, he asks the model to recreate W. H. Auden’s notoriously difficult 1941 course "Fate and the Individual in European Literature" as an engaging learning experience; Fable 5.2 compiles syllabus materials, annotates 35 works with Auden’s scholarship, and publishes an open-source course website with critical apparatus.
Mollick notes persistent "jaggedness"—gaps where the AI falls short of human capability—particularly in judgment and taste. For example, early versions of the humanities site were either overly academic or childish until refined with feedback. The model also failed to flag its own issues with literary analysis until prompted. He concludes that while Fable 5.2 approaches expert-level performance in narrow domains, it is not yet capable of independently running a business or reliably exercising human-like aesthetic and critical judgment.
The author emphasizes that these demonstrations show progress toward general intelligence but cautions against equating current capabilities with full AGI, especially in open-ended, real-world contexts.
- Nov 10, 2026 · Wired
Kora, a Linux-based desktop OS from Cyanogen cofounder, targets AI power users with local model integration
Trust76 - Nov 9, 2026 · The Verge — AI
Former DOJ antitrust chief launches AI legal-research startup to challenge monopolies
Trust72 - Nov 8, 2026 · The Verge — AI
Rights holders of Elvis Presley and Eminem sue Suno over AI music clones
Trust75