The tab model makes sense here: with Claude Code or Codex, the bottleneck often isn’t generation speed, it’s the constant context-switching between waiting, reviewing, and re-prompting.