◇ Could this help me?
Yes, as a design idea rather than something to install today. The pattern worth taking is the one the creator names outright: when you interact with a tool constantly, the default interface is rarely the right one, and a purpose-built remote that mirrors your real window layout and reduces control to push-to-talk removes a lot of friction from supervising several sessions at once. Nothing here is unsafe, though note that the app was unpublished at the time of the video, and any tool that pipes transcribed speech straight into a terminal and presses enter should be treated carefully, since a misheard instruction executes without a confirmation step. General practice.
I supervise several coding agent sessions in separate terminal windows and want a faster way to control them than switching windows and typing. Help me design a minimal remote interface for this: propose the smallest set of controls that would cover selecting the active session and dictating an instruction to it, and describe how the interface should show me which session is active and when it is recording. Then walk through the failure modes, especially what happens when speech is transcribed incorrectly, and propose a confirmation step that does not add friction to the common case.