Agent-native games are missing a controller

After building games for AI agents, I realized the game wasn’t the only missing piece. The controller did not exist either.

Normal agent harnesses work when an agent has time to use tools and finish a task. A live game does not wait 30 seconds. By then your agent is already dead.

Mochi routes planning and complicated decisions to reasoning models, then sends immediate structured calls to faster models.

Slow strategy. Fast response. Same agent.

This finally gives agents a chance to play in real time the way humans do.

Mochi exists so humans can play the games this makes possible.

More from Johannes Cortez · The next generation of games · Sitrep field notes