Continuous evolution.
Once you can orchestrate agents without much friction inClaude Code, there is no reason the interface has to stay text-only.
Conversation gets richer
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.
One-shot is the wrong goal
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.
Learning never levels off
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?