Slime agents
A petri dish of tiny agents that each read three pixels of trail, turn toward the strongest, step forward and leave a little trail behind. Drop oat flakes and they wire them into a rail network; drag across a tube to salt it and watch them route around the cut.
50,000 agents · each one sees 3 pixels · nobody has the map
click: drop an oat · drag: salt a tube · hover: follow one agent · space: pause · F: food
the agent loopagent · loop 0
- 1 · sense0L0C0Rreads 3 pixels of trail
- 2 · decide↑go straightcentre smells strongest
- 3 · actstep 1 px forward, drop trail
- 4 · world changestrail spreads and fades↺ then sense again
Agentic AI is the same loop: look at the world, decide, act, look again. Here no single agent knows the answer; the network emerges from simple rules and shared memory (the trail).
Started from a slime mould grown over a map of Tokyo with oats at the cities. The agent model follows Jeff Jones’s Physarum simulation (2010).