OpenAI Ships Dots Always-on Agent Runtime

OpenAI published the Dots specification for persistent agent processes that maintain state across sessions and execute background tasks without user prompts. The announcement lists integration hooks for calendar, email, and device sensors plus a policy layer for permissioned actions. No public benchmark numbers or latency figures appear in the post.

Related work includes the 2023 ReAct paper on interleaved reasoning and action loops and Anthropic's 2024 computer-use agent release, both of which documented failure modes when agents retain long-running context. Dots documentation does not reference these error patterns or provide equivalent evaluation harnesses.

Operational deployment will require new access-control primitives on host systems and audit logs for every background action. Enterprises already running persistent automation report that agent drift and credential sprawl emerge within weeks when state is not externally verifiable.

Next milestone is the claimed public API release; absent a CVE or incident report tied to Dots, adoption signals will appear first in developer telemetry from the OpenAI platform rather than enterprise case studies.