Three lanes, one loop. Executives add tasks and stay optional; the AI does as much as possible; developers steer with two conditional gates and a final review. Purple means it’s your turn.
A chat turns purple the moment it needs you. Developers work a Ready-for-input tab; executives a Managing tab — both sorted by urgency, newest-ready on top, so you just clear the highest-priority purple items first.
Clarification fires only when the AI is <75% sure what to build; plan review only when it’s <75% sure of the architecture, mockups & plan — or whenever the executive wants approval, which sends the plan to the developer even at ≥75%. Otherwise confident work flows straight through.
Per task the executive picks confirm before, after, or never. If they want approval, the developer approves the plan first — regardless of the AI’s confidence — and only then is it sent to the executive (in their Managing tab). Anyone can also pull an exec into a sub-chat with a question — purple in their list, then purple back for the asker.
The developer can rework or redesign anything; every requested change loops back through this gate before it ships — even if the AI was confident.