Chess of Minds

4 min read Original article ↗
Chess of Minds

Chess · reimagined

Chess, played by
twelve minds.

Six AI agents per side, one per piece-type, each with their own personality, voice, and opinions about the position. They propose, argue, and negotiate every move. The point isn’t to play perfect chess. It’s to watch minds collide.

Paste this into Claude Code

> Use https://chessminds.fun /play.md to start a chess game

Your Claude Code session does all the AI work. No API keys, no accounts. Free for Pro/Max subscribers at the token level.

Italian Game after 4.Bc4Imagine six voices arguing about what white plays next.

Why this exists

An experiment in watching AI agents decide.

01

Multi-agent coordination, as spectacle.

Chess is minds colliding. We split one mind into six per side and watched them try to agree. The arguments are more interesting than the chess.

02

Strategies are a design space.

Auction, democracy, monarchy, debate, consensus, hierarchy. Each produces a different game from the same position. This is collective-decision theory you can watch in 60 seconds.

03

Agent-native UX.

You point your coding-agent at one markdown URL and it runs the whole game locally. No accounts, no keys, no install. It’s a demonstration of what apps look like when agents are the runtime.

One turn, unpacked

How the twelve minds decide each move.

Chess of Minds replaces one player with six AI agents, each responsible for one piece-type (pawns, knights, bishops, rooks, queen, king). Every turn they propose, argue, and negotiate a single move. Here’s the loop.

  1. 01Brief
    Board → six inboxes

    The current position is sent to every agent on the side to move, along with the list of legal moves for their pieces only.

  2. 02Propose
    Six speeches, one round
    • ♟Pawnse2e488
    • ♞Knightsg1f374
    • ♝Bishopsf1c462
    • ♜Rooksa2a318
    • + Queen, King

    Each agent returns a single move plus a confidence score and an in-character speech. They also sometimes trash-talk the opponent.

  3. 03Negotiate
    Auction · Debate · Vote · …
    • ♟88
    • ♞74
    • ♝62
    • ♜18
    • ♛35
    • ♚4

    The chosen strategy picks a winner. In auction, the highest confidence wins. Other strategies use votes, debate rounds, or hierarchy. Different strategies make very different games.

  4. 04Play
    One move, server-validated

    The winning move is played against a real chess engine (python-chess). If it was illegal, the agent is asked to try again. Then the opposing six do the whole thing in reply.

Two sides × one deliberation each = one full turn. Repeat until checkmate, stalemate, or you hit the turn cap.

The cast

Six agents per side.

Each piece-type is one agent. The Pawns agent speaks for all eight pawns at once; the Knights agent for both knights. Their personalities come from the preset. This is medieval serious, the default. Other presets (Shakespearean tragedy, modern office) reshape every voice.

  • ♚

    KingCautious · self-preserving

    Prefers defensive moves. Knows that their death ends the game. In monarchy mode, they pick the winner from the others' proposals.

  • ♛

    QueenDecisive · often overconfident

    The most powerful piece, and they know it. Tend to propose sharp attacking moves with high confidence, sometimes prematurely.

  • ♜

    RooksStoic · structural

    See the game in ranks and files. Advocate for open files, doubled rooks, and the seventh rank. Rarely speak first.

  • ♝

    BishopsScheming · patient

    Diagonal thinkers. Fond of long-term pins, fianchettos, and quietly menacing moves. Often carry grudges across turns.

  • ♞

    KnightsGlory-seeking · tactical

    Jumpy, optimistic, see forks where none exist. Propose aggressive outposts and sacrificial charges with high confidence.

  • ♟

    PawnsLoyal · forward-marching

    Eight voices, one agent. Strongly biased toward pushing pawns. Speak collectively; know one of them will be sacrificed soon.

Strategies

How the winner gets picked.

The negotiation strategy is the most important choice you make. Same agents, same personalities, same board, but a different strategy produces a different game. These map to real collective-decision mechanisms, which is part of why watching them is interesting.

  • auctionstrategy

    Highest confidence wins.

    Fast, punchy turns. The pieces with the strongest conviction win, which rewards bravado as much as chess skill. A good default.

  • democracystrategy

    One vote per agent. Plurality wins.

    Coalition games emerge. The pawns are numerous but vote as one. The queen is one voice among six. Ties break by auction.

  • monarchystrategy

    The King picks the winner.

    A second decision round: the King reads all teammates' proposals and picks one (or overrides with a move of their own). Watch the power balance.

  • debatestrategy

    N rounds of revision, then auction.

    After the first round, each agent sees the others' proposals and may change their mind. Minds really do change, which is the point.

  • consensusstrategy

    Keep debating until ≥75% agree.

    Slow, collaborative, occasionally gridlocked. Falls back to auction if it times out. Produces the calmest games and the worst blitz.

  • hierarchystrategy

    Queen > Rooks > Bishops > Knights > Pawns.

    Top-down: if the queen proposed, the queen wins. If not, the highest-ranking proposer. Deliberate authoritarianism, sometimes brilliant.

  • rotatingstrategy

    A different piece-type has sole authority each turn.

    Turn 1: pawns decide. Turn 2: knights. Turn 3: bishops. Keeps no agent from dominating; produces wildly uneven games.

  • anarchystrategy

    A random proposal wins.

    Pure comedy. Useful as a control: lets you see how much the strategy choice was doing in the first place.