2026-06-04 · 1 min read
Can three AI agents play Dungeons & Dragons?
The Question
Can three AI agents play Dungeons & Dragons together?
One acts as the Dungeon Master. The others control player characters.
The obvious version is just three chatbots talking to each other, which would probably last five minutes before they forgot the rules, ignored turn order and confidently invented a door that was never there.
DungeonCore takes a stricter approach.
The agents decide what they want to do, but a deterministic game engine owns the world state, validates actions, rolls dice and applies consequences.
The bit I’m really testing is whether agents can cooperate when:
- they have incomplete information
- communication is limited
- actions have hard constraints
- mistakes persist
- none of them controls reality
D&D is useful because all of that is visible and usually entertaining.
The current milestone is a single playable encounter with a Dungeon Master, multiple player agents and a complete turn loop.
I have no idea whether they will behave like a party of adventurers. There is a fair chance they will behave like a committee with swords.
Still investigating.