Animation compares an LLM fly with a Jev fly dodging a swatter
A fully code-generated animation made with Opus 5.5 shows Jev returning ESCAPE, RIGHT, 0.97 in 100 ms while an LLM agent is still writing its sentence.
Jev + Opus 5.5. Based on a true story. Pasadena, California. Poor fly. Two AI flies. Same body. Same table. Different brains. The left one runs like an LLM agent. It thinks in words. It has to write before it moves. The right one runs like Jev. It doesn't write. It returns one decision: an action, a direction, a confidence score. 0 ms. The swatter's shadow hits the table. 100 ms. Jev returns ESCAPE, RIGHT, 0.97. 150 ms. The fly crouches, raises its wings and jumps. 215 ms. The swatter lands. The LLM fly is still thinking. It finishes its sentence seconds later. "I should consider moving to the left, because a large object is approaching." The scene is an illustration. These two numbers are real. Caltech clocked it in 2008: a real fruit fly plans its escape from a swatter in about 200 ms. TypeSafe says Jev decides in about 100. To click a button, your agent needs one decision. Every frame is code. Built with Claude Opus 5.5. No Blender. No 3D artist. Which brain is your AI agent running on?
This post quotes another post: “300 agents build a fruit fly part by part, then Jev takes over as the brain”