The era of the chatbot that forgets you the moment you close the tab may be ending. At its DevDay developer conference last month, OpenAI unveiled "dots" — always-on AI agents in ChatGPT designed to keep working toward a user's goals without being told every step. "Dots are remarkably capable, always-on agents built to handle really anything you can think of," CEO Sam Altman said during the keynote.
Each dot runs on GPT-6 Astra with its own cloud computer and browser, separate from the user's devices and data unless the user chooses to connect them. You give a dot a goal, connect the relevant applications, and set rules about what it may do independently and what needs approval. The dots connect with more than 4,000 applications through OpenAI's plugin ecosystem, and users can reach them through ChatGPT's web, mobile, and desktop interfaces, plus Slack and Microsoft Teams, with text messaging support planned.
The concept shifts AI from answering questions to owning responsibilities. OpenAI's examples include investigating software bugs as they surface in Slack, preparing an invoice a user forgot to send, spinning up a working app the moment a new design file lands, and revising research analysis as new data arrives. When users are not interacting with their dots, the agents can conduct what OpenAI calls "proactive research" through connected applications — restricted to read-only access so they cannot send messages, change content, or control the user's computer.
That autonomy comes with guardrails. Users define Custom Rules specifying which actions are allowed, blocked, or require explicit approval, and an automated review system can pause or stop a dot mid-task if it detects a safety concern. The clearest line: sensitive actions such as changing a password always stay with the human. OpenAI also published a dedicated safety write-up alongside the launch and cautioned, "Dots can still make mistakes, so always review consequential work."
Dots roll out first to ChatGPT Pro subscribers — the tiers run $100, $200, or $500 a month in the U.S. — along with Business Premium and Enterprise workspaces, for users over 18 outside the EEA, the UK, and Switzerland. Your first dot is included in the plan; additional dots and capacity are coming later, pricing to be announced. The old Agent mode is being retired in favor of this persistent approach. Between dots, the new $500-per-month Pro 500 tier, GPT-6.1 Sol, and a shared ChatGPT Space workspace, DevDay was OpenAI's clearest signal yet: the future it is selling is not a smarter chat window, but a staff of tireless digital workers.