Analysis
Ethan Mollick calls Claude Fable 5.2 “a sort of Jagged AGI” after early access: near-expert on many computer tasks, but weak at judging its own work; we read the reports, not his post
After early access, Ethan Mollick calls Fable 5.2 “a sort of Jagged AGI”: near-expert on many computer tasks, weaker at judging its own work.
Wharton professor Ethan Mollick, who had early access to Claude Fable 5.2, wrote on November 9, 2026 that it “crosses the line into being a sort of Jagged AGI”. In his post, “Fable 5.2 and Jagged AGI”, he describes near-expert work across many computer-based tasks, alongside a weakness: the model struggles to judge the quality of its own output. We could not open the post itself, so this summary relies on reports that quote or paraphrase it.
The essentials
- His definition: SFEIR says Mollick defines AGI as AI as good as most human experts at most intellectual tasks done on a computer, and judges Fable 5.2 “almost” there. Brave New Coin notes that this definition leaves out many human tasks and most jobs.
- What he tried: GPTunneL and SFEIR relay tests including writing games from a keyboard app’s code, picking a small business idea, and a research paper on company-registration costs that grew to 182 pages with shortcomings. A portal-worlds project reached 450-plus worlds from about 25,000 agent launches, per the model’s own estimate.
- The limit he found: Techmeme summarizes it as a lack of self-directed judgment and taste. GPTunneL puts it as limited ability to assess the quality of its own work. Brave New Coin quotes “The AI had built a bureaucracy” about agent teams that produced an 87,500-word rulebook that slowed a project.
- Disclosures: Brave New Coin says Mollick used free prerelease access, takes no money from AI labs and has no financial interest in Anthropic’s IPO. He also stopped a business experiment at the demonstration stage, so it did not show the AI could run a business alone.
Reading the label
“Jagged” means uneven: strong on some tasks, weak on others. Mollick’s claim is bounded by his own definition, which covers computer-based intellectual tasks, and by his stated gap in self-judgment. The French summary “expert-level on some tasks, no autonomous judgment” compresses it: reports say near-expert on many tasks and limited, not absent, self-assessment. One commentator, @matdryhurst, doubted that the 3D-worlds demo shows advanced creativity, per Techmeme.
What this is not
It is one tester’s hands-on account with free access, not a controlled study. His other point, that Fable 5.2 sits above trend on several benchmarks, comes from his X posts as relayed. We did not read the post. Not investment advice.
What is not established
- Mollick’s exact wording and caveats: not read directly; relayed by five outlets.
- A controlled comparison of Fable 5.2 against human experts: none found.
- Whether Fable 5.2 can run the business idea alone: not shown; he stopped at a demonstration.
- Agreement among experts that this counts as AGI: none found; the label is Mollick’s own.
Summary table
| Point | What is stated as of November 7, 2026 | Source |
|---|---|---|
| Label | “A sort of Jagged AGI”, posted on X and in a blog post on November 9 | Techmeme |
| Definition | As good as most experts at most computer-based intellectual tasks | SFEIR |
| Weakness | Limited ability to assess the quality of its own work | GPTunneL |
| Example | Agent teams wrote an 87,500-word rulebook that slowed a project | Brave New Coin |
| Independent score | ARC-AGI-3: 81.2% in the standard harness, per a newsletter relay | Handy AI |
FAQ
Does Mollick say Fable 5.2 is AGI?
He says it is “a sort of” Jagged AGI under a narrow definition, near-expert at many computer tasks but weak at judging its own work.
Was he paid by Anthropic?
Brave New Coin says he had free prerelease access, takes no money from AI labs and has no financial interest in Anthropic’s IPO.
Did you read his post?
No. We could not open it and relied on outlets that quote or paraphrase it.
Method and limits
Sources checked on November 10, 2026: Techmeme, Brave New Coin, GPTunneL, SFEIR and Handy AI, read as search extracts; Mollick’s post and posts on X were not read directly. Not investment advice. Disclosure: OrigoTech drafts with Claude, a model made by Anthropic, the subject of this article.
Sources
- Techmeme: Anthropic's Fable 5.2 represents “Jagged AGI” (Ethan Mollick / One Useful Thing) (verified 2026-11-08)
- Brave New Coin: Claude Fable 5.2 and Mythos 5.2: Benchmarks and Jagged AGI (verified 2026-11-08)
- GPTunneL: Claude Fable 5.2: Review, Benchmarks and Reactions (verified 2026-11-08)
- SFEIR: Claude Fable 5.2 et Mythos 5.2 : prix, benchmarks, coût réel (verified 2026-11-08)
- Handy AI: Model Drop: Fable 5.2 & Mythos 5.2 (verified 2026-11-08)
Spotted an error in this article? Report it · Our corrections policy