OpenAI's 'Dots' Are Always-On Personal AI Agents Built to Act Without Being Asked

OpenAI is moving decisively beyond the chatbot model with a product called Dots — proactive, always-on personal AI agents embedded in ChatGPT that can find ways to help users without waiting for a prompt. Alexander Embiricos, the product lead for Dots, spoke publicly about the initiative at OpenAI's 2026 DevDay conference, laying out the company's vision for what comes after the conversational AI era.

Dots are designed with their own dedicated compute environments and can maintain context across the different surfaces where users interact with them. A key design philosophy is that most people will only ever need one personal agent — a single Dot that knows them well enough to handle the full range of their needs. OpenAI is also building out specialist Dots that companies can deploy and manage centrally for shared team workflows, expanding the concept beyond individual use. The company is also launching Space, a collaborative workspace where both humans and agents can work together on documents, data, and ongoing projects.

One of the more revealing details: OpenAI started Dots with deliberately conservative permission defaults, requiring user confirmation before taking actions. Early testers — including the interviewer — found it over-cautious, with the agent repeatedly asking for approval even for mundane tasks like buying shoes. OpenAI has since dialed back unnecessary confirmations based on that feedback. The other striking anecdote came from inside the company itself: an employee reportedly woke up to find their Dot had autonomously identified and fixed a software bug overnight.

On pricing, Embiricos frames Dots like a contractor who is always available but scales their effort based on what you're paying. The model envisions a flat monthly fee, with the agent potentially shifting to cheaper underlying models or deferring lower-priority tasks to off-peak hours as users approach usage limits. Notably, Dots currently run on OpenAI's Astra model with unmetered access that doesn't count against existing ChatGPT plan limits — a temporary arrangement as the product matures. OpenAI is explicitly ruling out taking transaction cuts from purchases made through Dots, instead betting on users doing high-value, ambitious work that justifies a premium subscription.

The longer-term roadmap includes opening Dots to free ChatGPT users and enabling cross-agent communication — letting one person's Dot interact directly with other people or even other Dots. OpenAI frames its competitive position against rivals like Meta's Muse and Google's Instinct on two axes: being the most capable agent and the most trustworthy one. How the company threads that needle — autonomy versus permission, ambition versus privacy — will define whether Dots becomes the dominant personal AI paradigm or a cautionary tale about agents that overstep.