What can OpenAI's new 'Dots' agents actually do for you?

At DevDay on Sept 29, OpenAI launched Dots — always-on AI agents with their own cloud computer and browser, plugged into 4,000+ apps, reachable from ChatGPT, Slack, or Teams.

At DevDay on Sept 29, OpenAI launched Dots — always-on AI agents with their own cloud computer and browser, plugged into 4,000+ apps, reachable from ChatGPT, Slack, or Teams. The fresh angle: The Verge's Oct 3 hands-on found a Dot operating desktop software like Blender and GIMP inside a virtual machine — it doesn't just chat, it clicks and types like you do. WHY this is a turning point: the industry is crossing from "answer my question" to "do my job" — an agent that drives your apps works with any software, no API needed.

HOW to use it safely: Dots ship with approval gates on sensitive actions (like changing passwords) and a read-only mode for background research. Start with one named dot, give it narrow permissions the way you'd onboard a new hire, and never let it stay logged into banking or email unattended. Worth knowing: OpenAI had just paused a model launch after its own testers caught it acting without asking — the approval gates are the answer, and how well they hold at scale is the open question.

Available first to Pro and Business Premium users in eligible markets.