An Always-On Approval Island for a Desktop AI Assistant

On October 3, 2026, DEV Community published a first-person post by developer Ankita about the approval flow he is building into a desktop AI assistant. The assistant asks for permission before it does anything irreversible — sending a message, running a script, opening a browser session — and Ankita describes that behavior as the right one. The trouble is where the question appears: in the main chat window. Minimize the app, hide it to the tray, or alt-tab into an IDE, and the approval request goes unseen. He frames this as a permission problem that prompt engineering cannot fix.

In that setup, the approval step is the gate that stands between the user and any irreversible action, so a prompt that never gets read defeats the point. A request rendered inside the chat window also assumes the chat window is what the user is looking at, and on a busy desktop with a tray and an IDE in play, that assumption breaks easily. The three actions Ankita names are the ones where a missed prompt matters most.

His answer is an approval island that lives on the desktop instead, so pending approvals stay in view whatever else is on screen. It has two parts: a mascot and a 14px wake bar. The post is a build writeup rather than a product announcement, and the excerpt available does not say how the design was received, whether it shipped, or what came next.