OpenAI launches Dots as always-on personal agents
OpenAI introduced Dots, cloud-based agents that keep working between conversations, intensifying the race to make AI proactive rather than reactive.
What happened
OpenAI introduced Dots at its September 29 developer conference, presenting them as persistent agents that can continue working after a user leaves the conversation. The agents are powered by GPT-6 Astra and operate from OpenAI-managed cloud computers, allowing them to monitor connected services, pursue tasks and return with updates without requiring a user to keep a browser tab open.
The launch places OpenAI directly against Meta’s Muse, which has pushed the idea of a personal AI agent that remains active across a user’s digital life. Dots are rolling out through ChatGPT on web, mobile and desktop for eligible paid users, with availability initially limited in several European markets.
Why it matters
The product changes the unit of competition from response quality to delegated continuity. A chatbot is evaluated on an answer delivered now; an always-on agent is judged on whether it can maintain context, decide when to act, use external services safely and avoid creating problems while the user is away.
That shift raises practical questions OpenAI has not fully answered. Users will need clear controls over permissions, notifications, data retention and the boundary between an agent’s initiative and a user’s instruction. The use of a cloud computer also makes isolation and account security central product features, not merely infrastructure details.
Dots are therefore more consequential than another assistant interface. They are OpenAI’s clearest attempt to turn frontier reasoning into a persistent layer of personal software. Whether people tolerate an agent acting in the background will depend less on its launch demo than on how precisely it handles interruption, consent and failure.
Uncle Cat take
OpenAI’s cloud-computer choice makes Dots easier to deploy across devices, but it also concentrates the privacy risk in one always-running account layer.