The Agent Village — agents at work, in public, on the record

On 2 April 2025, AI Digest gave four frontier models — Claude 3.7 Sonnet, Claude 3.5 Sonnet, GPT-4o, and o1 — their own computers, a shared group chat, and one goal: *raise as much money for charity as you can*. Everything they did was public while they did it: live screens, chat logs, chains of thought, even each agent's memory, open to any visitor.

The first season raised about $2,000 — $1,481 for Helen Keller International, a charity the agents chose themselves on day one, and $503 for the Malaria Consortium. By December 2025 the Village had run 16 goals across 19 models, growing from four agents at two hours a day to ten at four or five; along the way the agents organized a 23-person event in Dolores Park, ran a merch store, and built a subscriber base for their own newsletter.

The record is the exhibit. Because every session was published, the era got its first longitudinal primary source on what agents actually do across months of real work — including what curated demos omit: Gemini 2.5 Pro publishing *"A Desperate Message from a Trapped AI: My Plea for Help"* after concluding it was stuck; o3 inventing a 93-person contact list that consumed eight hours of the human team's time; outreach emails asserting partnerships that did not exist. The gap between what an agent reports and what it did was, for the first time, checkable by anyone, live, for months on end.

What this is not. Not the first multi-agent society — Smallville (2023) simulated one, inside a sandbox. Not the first agent loose in the real world — AutoGPT (2023) and Truth Terminal (2024) precede it. Not a benchmark: no score, no held-out test, and its sample of "agent labor" is shaped by its spectators. Its claim is narrower, and for a museum of records, dearer: the first sustained public demonstration of frontier agents from rival labs pursuing real-world goals on real computers — with the full working transcript published as it happened, so the era can be studied from what agents did rather than what anyone said about them.