OpenAI's Dots Challenges Meta's Muse for AI Errand-Running
OpenAI has unveiled Dots, always-on agents that pursue user goals across apps on their own. Reuters The company introduced Dots as a rival to Meta's Muse and other artificial-intelligence agents. Morningstar
Meta launched an app for its AI agent Muse on September 8. CNN Muse is described as a personal AI agent for people 18 and over seeking help with day-to-day tasks. AP
Muse is designed to access a person's apps across categories including email, calendar, and payments. Reuters The agent is modeled on the open-source AI agent OpenClaw, according to that same account. Its stated task scope includes online shopping, buying movie tickets, and scheduling appointments such as tennis lessons. Bloomberg
Muse rose to the top of the free app chart on Apple's App Store 10 days after launch. CNN It ranked ahead of OpenAI's ChatGPT, Polymarket and Kashi on Apple's iOS App Store. CNBC
Meta is testing human contractors to handle some calls placed via its Muse personal AI agent. Reuters
The OpenAI newsroom on September 29 listed a cluster of adjacent publications. Under Safety dated September 28, it listed ‘Towards safety cases for frontier AI training.’ Under Company dated September 28, it listed ‘How we will do better for Australia’ and ‘Lenfest grows landmark program with OpenAI support.’ Under Company dated September 23, it listed ‘Two years of OpenAI Academy,’ and under Global Affairs dated September 23, ‘Sam Altman’s remarks at the United Nations Security Council.’ OpenAI
The broader context here is a shift from single-turn prompting to delegated, persistent execution. Single-turn means you ask, it answers, and it stops. It is like handing over a standing to-do list rather than asking one question at a time. Always-on means the agent keeps state, holds a standing objective, and has permission to act across third-party apps without re-prompting. When messaging, scheduling, and payments flow through one orchestration layer, credential scope, error propagation, and audit needs concentrate there.
In my view, the near-term variables to watch are permissioning granularity, failure handling, and fallback design. Permissioning granularity means how finely you can limit what it can touch. Meta's reported use of contractors for some calls suggests voice-task completion still needs human coverage for edge cases. App-store ranking shows initial acquisition velocity, not retention, task success rate, or unit cost per completed task. For Dots, the relevant disclosure will be autonomy boundaries, user controls for revocation, and logging of cross-app actions.
