Scaling reinforcement learning, tracing circuits, and the path to fully autonomous agents