OpenAI has shipped 'dots,' a new agent layer inside ChatGPT that runs tasks autonomously in the background without requiring the user to stay in the conversation. Each dot is a persistent, proactive assistant tied to a specific project or recurring task, continuing work after the chat window closes.
The meaningful architectural shift here is persistence. Unlike standard ChatGPT sessions that reset, dots maintain state across time and pick up where they left off. OpenAI frames user control as a core design constraint, meaning dots are built to act but also to pause and surface decisions back to the user at defined checkpoints.
The blog post is worth reading for the specifics on how task handoff and user oversight are actually implemented, and what kinds of tasks OpenAI considers in scope at launch. The control model is the thing to scrutinize: autonomous agents are only as useful as the guardrails that prevent them from running in the wrong direction.
[READ ORIGINAL →]