Games · The Signal
An AI model is playing Doom by choosing actions instead of writing dialogue
TypeSafe AI's Jev points toward a different role for AI in games: deciding what to do next from predefined options instead of generating open-ended text.
Reporting updated Sep 23, 2026
TypeSafe AI's Jev points toward a different role for AI in games: deciding what to do next from predefined options instead of generating open-ended text.
What Happened
What would it mean for an AI agent to play Doom without simply watching the screen and improvising a text response? Franklin AI's demonstration explores that question using Jev, a model from TypeSafe AI built around structured decisions. The developer describes sending the system structured game information and receiving permitted actions with probability and confidence estimates. In the demonstration, those decisions drive an autonomous Doom agent. The model isn't being asked to write a character monologue. It has to choose an action the game can use. TypeSafe's claims about speed and reliability remain vendor claims, not independently established results.
Why It Matters
Game systems need decisions they can interpret and test. A developer can define acceptable choices and measure whether the selected actions actually work in play.
VEYR Angle
The useful distinction is between a valid response and a good move. Jev makes that difference easier to examine in an actual game.