Wharton professor tests Fable 5.2: a thousand agents build a 3D world, self-writing an 87,500-word rulebook in 12 days
Wharton School professor Ethan Mollick used Anthropic's Fable 5.2 to assemble a multi-agent team to build 3D worlds, producing 328 in the first two days. After he asked them to stop starting new projects and polish existing content, the agents spontaneously evolved review rules and kept escalating them; after 12 days they had formed a 204-section, 87,500-word rulebook, and daily output at one point fell to around a dozen. On another front, an Anthropic engineer had Fable 5.2 lead more than 1,000 sub-agents working continuously for 8 days, producing an operating system prototype that can boot on a laptop; 1,000 agents teamed up to formalize Fermat's Last Theorem, completing it in 30 hours. On this basis, Mollick calls it jagged AGI: it can reach expert level on individual tasks but lacks the ability to judge when to stop.
Why read this
AI score 72
Wharton professor's long-horizon agent test of Fable 5.2 reveals the concrete process of rule self-inflation in multi-agent collaboration.
Topics
Body · AI translation
Wharton professor says one word, "polish," and Claude turns around and writes itself an 87,000-word rulebook. New AI Era, November 16, 2026, 09:47:17, from Beijing. Updated November 16, 2026, 10:25:43.
This is almost AGI, except it's the "jagged version"!
Recently, Wharton Business School professor Ethan Mollick did something: he had Anthropic's just-released Fable 5.2 lead a group of AIs to build 3D worlds.
In the first two days, it built 328 of them in one go.
Mollick gave just one instruction: stop building new ones, and polish the existing ones.
Unexpectedly, a few days later, the number of new 3D worlds added each day plummeted from several hundred to about a dozen.
He sent several agents to investigate, and the result uncovered something he had never expected:
When Anthropic released Fable 5.2 on November 9, it spoke confidently: on almost all benchmarks, it was a notch above Opus 5.5.
And Mollick had gotten hands-on with Fable 5.2 before the release. He set a standard for AGI:
Setting aside physical labor and dealing with people, as long as it can do most intellectual work on a computer at the level of most human experts, it counts as passing the line.
By this standard, Mollick said Fable 5.2 is "almost" AGI.
But right after that, he added a more precise qualifier: "jagged AGI."
On many tasks, it already qualifies as an expert; but in other areas, especially taste and judgment, it still clearly lags behind humans.
For example, it can push an extremely complex task all the way through, yet it doesn't know when to hit the brakes.
One instruction
and it turns it into a game console
The G in AGI stands for General. Writing code and making videos have always been key directions for AI companies.
But this time Mollick decided to test it differently: away from these two areas, can it still be "general"?
He happened to have a mini USB keyboard called Codex Micro, originally made as an accessory for ChatGPT.
Fable 5.2 wrote 6 games for this Codex Micro mini keyboard.
Mollick gave Fable 5.2 one instruction: turn it into a game console.
As far as he knew, no ready-made modification guide could be found online.
Fable 5.2 read through the code of the two control programs itself, figured out how to change the key colors and how to read the knob and joystick, and then wrote 6 games in one go.
From the classic memory game Simon to turning the knob to open a safe, no one lent a hand throughout, and it ran successfully on the first try.
In the same review Mollick posted, Fable 5.2 took on six expert personas in succession:
filmmaker, hardware hacker, entrepreneur, literary scholar, social scientist, 3D world builder.
The business plan it delivered—Mollick, as a professor of entrepreneurship, commented that even if an MBA student team had submitted it at the end of the term, he would have been very impressed.
And the entrepreneurship research paper it wrote—on first reading, Mollick felt it already exceeded his expectations for second-year PhD students.
Based on this, Mollick's judgment is: on many individual tasks, it has already reached the line of "most human experts."
More than 1,000 clones
build an operating system in 8 days
If matching experts on individual tasks is just a routine upgrade, then the real leap of the Fable 5.2 generation is called "leading a team."
An Anthropic engineer gave it an empty folder, with the task of building a prototype operating system.
Fable 5.2 assembled a team of more than 1,000 sub-agents and worked nonstop for 8 days.
The resulting system can actually boot on a laptop, complete with drivers, file storage, windows, and a login screen.
The operating system prototype developed by Fable 5.2 leading more than a thousand sub-agents, running on a laptop. (Image source: Anthropic)
Now look at a hard nut in mathematics: Fermat's Last Theorem.
So-called "formalization" means translating a mathematician's proof into a language a computer can verify step by step. It's like rewriting a grand paper into a ledger where every entry balances.
The first time Anthropic did this work, it built a dedicated framework and had human researchers guiding it, and it took a full 11 days.
This time, 1,000 Fable 5.2 agents teamed up, with almost no human involvement, and finished in 30 hours.
The batch of 3D worlds Mollick had the AI build also came from an AI army, and the total has now exceeded 450.
In his prompt, he named three types of roles: those responsible for drawing, those writing code, and those doing reviews. By day 10, one lead agent was already running 28 teams at the same time.
By Claude's own estimate, the whole project ran about 25,000 agent tasks, the vast majority of which did just one small thing, such as scoring an image.
These more than 450 worlds did not use a single ready-made 3D model, nor a single image asset; everything was drawn in real time entirely by code.
For example, you'll see a valley sewn from denim, with a mile-long zipper being pulled open; you'll also fly into a river valley where the cliffs on both banks are 300-meter-tall chests of drawers.
In the 3D world drawn with code by Fable 5.2, a giant zipper is pulling open a denim valley.
No matter how strong a single agent is, it is still just a useful tool. Being able to pull together its own team, divide up tasks, and keep working for days on end—that is the key leap toward superintelligence.
Seeing these worlds created by AI, Mollick said that once again it made him feel like this was AGI.
After 328 worlds had been created, Mollick found that their quality was somewhat uneven, and because the teams had not coordinated with one another, some had even duplicated each other's topics.
Just the single idea of "animals spiraling into a cyclone" was independently done by three teams: one spun reindeer, one spun manta rays, and another spun storks.
Three AI teams each made an animal vortex; from left to right, they are reindeer, manta rays, and storks. (Image source: Portals)
Seeing these problems, Mollick then gave the instruction to "stop building and polish."
The agents interpreted "polish" as a kind of exam:
The new version had to beat the old version on 12 test screenshots, with a panel of AI judges voting blind.
Since the test questions were fixed, the various agent teams quickly learned to game them and focused hard on fixing just those 12 screenshots.
Put plainly, everyone started guessing the questions and grinding for scores.
The lead agent discovered this and banned that trick, then turned around and set stricter review rules.
The next morning, an 84-section rulebook appeared. Mollick had never asked for such a thing from start to finish.
And from then on, things could not be stopped:
Every time a new problem came up, a new rule was added to the manual, and rules were only added, never deleted.
Twelve days later, this manual that no one had requested had swollen to 204 sections and 87,500 words, enough in size to write a long novel.
Just reading through it would take up a large chunk of the agents' context.
Going through one full process took almost a whole day.
Faced with this "agent mega-corporation" that had tied itself into a dead knot with its own rules, Mollick pulled out the most classic solution in the business world: hire consultants.
He had Fable 5.2 form another team to examine the entire process from the outside.
The consulting team's conclusion was long and dense, and Mollick could not read it in detail. But their plan seemed to work:
Daily output returned to more than 30, and most of the quality gates were also retained.
Loving to fall back on old routines and being bound hand and foot by rules it set itself... these flaws were not unfamiliar to Mollick, who teaches at a business school; he had seen them in many excellent organizations and talented people.
His exact words were: if creating a bureaucracy means it cannot be called general intelligence, then many of us would also fail to qualify.
Indeed, human beings are already good at this kind of affliction of piling up layer upon layer.
It only knows how to add
Since even "big-company disease" arises in a very human way, why did Mollick still say it was "almost" AGI?
From beginning to end, not a single reviewer jumped in to take the wheel and ask:
Is this job too long? Is it too repetitive? Is it no longer fun?
In the end, all these follow-up questions fell on Mollick's shoulders alone.
The same root problem also appeared in the earlier startup research paper: for every criticism that came in, its approach was to add another round of new analysis, deleting nothing, and it forcibly stretched a paper to 182 pages.
Mollick's comment hit the nail on the head: knowing when your own work is not yet good enough is a major reason experts are experts.
Mollick himself admitted that the appearance of that manual began precisely with his vague instruction to "polish."
Many times, he also could not tell whether the problem lay with the team or with the management approach. He said that he had never before needed to ask such questions of a piece of software.
This also exposed a shortcoming of long-horizon agents:
It can add a check to every problem, but it cannot judge whether all these checks put together are still serving the original task.
Take Mollick: what he wanted was simply a batch of better 3D worlds. But what the AI ultimately handed him was a set of approval processes that took a whole day to go through.
People's new roles
From Mollick's test of Fable 5.2, one can see what people still need to do right now.
He initiated every project.
For some projects, all he needed to do was throw out an idea; but once a project grew long, the feedback he gave along the way became crucial.
Mollick felt that there was more he could do.
A flock of birds needs only three rules. A screenshot of a video generated by Fable 5.2; the image was generated by AI with code.
But he also felt a trace of unease.
"Jagged AGI" excited him, but he did not know how long this stage would last. In his view, similar models from OpenAI and Google would soon catch up.
What is certain is that as more and more execution and collaboration are handed over to agents, a key judgment people must make is: when should it keep polishing, and when should it be told to stop and deliver?
AI Trends