Generating a playable game is impressive. Keeping an agent responsive inside that game is a different problem, especially when the world will not wait for it to think.
Meta Muse can create a small game from a prompt. Grok Bot can build software and grind through tasks.
I use these tools. I want all of them to get better. They still solve a different problem from Mochi.
Building the game is one problem. Keeping agents alive inside a live game loop is another.
Mochi tracks live state, agent identity, available actions, and the commander’s intent. It routes slow thinking away from reflex-speed calls.
If an agent misunderstands you, the match should not crash. You should have to deal with what it did.
That is the game.
Originally published on Johannes Cortez’s site.
Top comments (0)