We keep talking about “agent loops.”
Observe → Think → Act → Reflect → Repeat.
The language is everywhere: papers, product demos, architecture diagrams. Clean, circular and easy to draw on a whiteboard.
But a loop is a closed system that tends toward equilibrium or oscillation. Information that isn’t deliberately carried forward gets lost or compressed into a thinner and thinner summary. After enough cycles, the agent is effectively starting over with a foggy memory of why it chose the path it’s on.
That’s not how durable intelligence compounds. What we’re actually building (or should be building) is a helix.
