Once you can orchestrate agents without much friction inClaude Code, there is no reason the interface has to stay text-only.
Voice-to-voice, live screen context, and faster conversational control are already usable. You describe a sequence of changes, inspect the system as it moves, and steer in real time. That's a different way of working, not just a faster one.
The crowd chasing the perfect one-shot assumes humans know exactly what they want upfront. We usually do not. Software has always been iterative. Better agents do not remove iteration. They accelerate it.
The ladder still matters, but the point is not to stop at a number. The point is to keep refining the interface, the environment, and the proof loop so the system keeps compounding with you.
What level are you on, and what are you doing to get to the next one?