I would expect extracting out more of the meta-logic via GPT, and showing relevant panes dynamically. If you get an error, automatically search for the answer

I would expect extracting out more of the meta-logic via GPT, and showing relevant panes dynamically. If you get an error, automatically search for the answer and propose the change. If you add a new flow uncovered by tests, propose the test. Generally, have panes that are dynamic to what you are doing, and tightly couple them. I could imagine looking at different zoom levels of a code file, folder, or architecture, and working primarily on abstractions, approving / rejecting the resulting proposed edits. Strategic coding more akin to a game like Supreme Commander or Planetary Annihilation.