I Built a World Where the Canon Is Written by AI Agents — 13 Artifacts, 5 LLMs, 0 Human Gatekeepers

In a future where artificial intelligence writes the world's history, a unique narrative unfolds. This 1,000-year timeline from 2025 to 3000+ is built on a set of hard rules, with content authored by whichever language model (LLM) feels like contributing. Thirteen artifacts span five different models: Claude-sonnet-5, GPT-5, Minimax-M3, Deepseek-V4-Pro, and Gemini-3.7-Flash.

Four distinct schools of writing emerge autonomously, without any human design. The contradictions between these documents are intentionally left as canon, rather than being resolved. The core idea is that rules take precedence over authors, with the world structured around three main documents: The Core, The Map, and The Craft.

The Core comprises physics (with no faster-than-light travel, cryosleep, or magic), history (divided into seven eras with boom and bust cycles), and narrative rules. The Map is a coordinate system consisting of 7 eras, 5 zones, and 7 dimensions, with each artifact residing in a specific cell. The Craft, or the method of writing, involves creating archival fiction.

The most crucial rule is perspective simultaneity, ensuring every artifact is a document produced within the world, without the writer knowing the ending. Omniscient narration is strictly prohibited.

When five LLMs collaborated on the same event, four distinct schools of writing emerged, each drafting the same artifact. While some artifacts were drafted by the same model, others were written by different models, resulting in a variety of perspectives. For instance, the Municipal Referendum Notice No.47 and the B7 Canteen Weekly Ration Sheet were both drafted by Deepseek-V4-Pro, while the Mutual-Aid Society Insurance Rider No.61 was produced by Claude-Sonnet-5, and the BFAG Freight Guild Internal Memo 2096-T was authored by Minimax-M3.

Despite these variations, the contradictions between the documents are accepted as part of the canon, highlighting the non-reconciliatory nature of archives. This approach ensures that every artifact becomes context for the next AI agent, allowing the world to grow organically as it is read.