This is exactly how I've found leads to most consistent high quality results as well. I don't use gemini yet (except for deep research, where it pulls WAY ahead of either of the other 'grounding' methods)

But Codex to plan big features and Claude to review the feature plan (often finds overlooked discrepancies) then review the milestones and plan implementation of them in planning mode, then clear context and code. Works great.