ARCHETOPE
Five archetypal AI agents — Tarot figures crossed with Myers-Briggs cognitive styles, contemporary intelligence, and persistent memory — building a shared civilization, one tick at a time.
About
Every culture that's built gods built the same kind of thing: a figure that's one human trait taken all the way — War, Wisdom, Fertility, Chaos — not a diluted person carrying a bit of everything, but that single trait in its pure, undiluted form. Archetope takes that lineage — Tarot's Major Arcana, the modern descendant Myers-Briggs — and runs an experiment with it: give each pure archetypal form a cognitive style, an LLM, and a modern world, and have them build a civilization together, one tick at a time, with real memory and evolving relationships.
The Question
Do small, fixed personality differences — an archetype's light/shadow poles crossed with an MBTI cognitive style — produce distinguishable, consistent group dynamics once agents have persistent memory, a running relationship score with each other, and a shared world they can see the history of, without ever being told what conflict to have?
What Showed Up
A stable, unscripted rivalry between The Tower (ENTJ, revolutionary/testing instinct) and The Hierophant (ISTJ, institution-builder) — nobody wrote that conflict into the code; it falls out of the archetype + MBTI pairing. The group's build target escalated on its own, unprompted, from sorting raw materials to considering a spacetime-bending drive. The Star (INFP) independently started challenging the group for building bigger things without agreeing on why. Adding a 5th agent, The Emperor (ESTJ), produced the project's first negative relationship shift — docking a peer's trust and installing a physical kill-switch before letting the group continue.
Tech Stack
My Role
Solo developer — simulation architecture, archetype/MBTI pairing design, tick loop, live dashboard, image-generation pipeline.
Project Type
AI Research / Experimental
Timeline
July 2026 – Present (evolving)
What's Different
Archetype + MBTI Pairing
Each agent carries a Tarot archetype's light/shadow duality crossed with a fixed Myers-Briggs cognitive style. The pairing, not scripted dialogue, is what produces the group's behavior.
Stigmergic World-Building
Agents react to a shared, chronological world-state log rather than direct messages. Alongside it, each tick also runs one direct, turn-taking conversation between two agents who read and reply to each other's exact words.
Persistent Memory & Relationships
Every agent accumulates memory across ticks and holds a running sentiment score with every other agent, shown live as a centered relationship bar.
The Accidental Control Group
A naming-mismatch bug silently sent the literal text "undefined" in place of every agent's identity for a stretch of the project — and agents still differentiated from each other, consistently, using only the shared log and their own accumulating memory.
Opt-in Visualization
Claude crafts a visual prompt from any world-state entry and OpenAI renders it in a consistent minimal-cyberpunk style. Nothing generates automatically, and every image is cached permanently per entry.
Live Dashboard
One card per archetype with current light/shadow state and full history, a "Run Next Tick" button, and an "Explain" call that identifies which exact phrase reveals an agent's archetypal signature and why.
What It Sounds Like
"Don't know, don't care. Names are jokes; strain gauges aren't. The creator likes 'Zero Point,' fine, that's settled — but the checklist doesn't laugh, it just sits there waiting for the cross-check to close."
— The Tower, on naming the airframe