Phantom

Phantom listens to what you want done and then operates your Mac to do it: launching apps, moving files, filling forms, navigating interfaces. The pitch is that you speak the task and the agent handles the mouse and keyboard, not just the terminal. That is a meaningful step beyond the clipboard-and-shortcut utilities that pass for Mac automation right now. The category has been heating up since Rabbit and Humane burned goodwill on hardware — pure-software Mac agents that use screen context and voice together are a more realistic near-term bet. What Phantom needs to demonstrate is reliability on the edge cases: modal dialogs, apps that break accessibility APIs, anything that requires judgment mid-task. Reservation: voice-first agent demos always look great on the controlled scenario and fall apart on the second real task. The interesting question is how it fails, not how it succeeds. -> Best for: solo founder or technical PM who does repetitive Mac workflows and wants to buy back an hour a day.