OpenAI has launched Dots, an agent platform of always-on personal AI assistants that each run on their own cloud computer, unveiled at the company’s DevDay on September 29, 2026. A Dot can keep working on projects after the user leaves ChatGPT, connect with more than 4,000 apps through a plugin ecosystem, and reach into the user’s own computer — a launch that reviewers say turns ChatGPT from a chatbot into something closer to “Codex for non-developers”.
Key Facts
- OpenAI announced Dots at DevDay on September 29, 2026 — always-on AI agents powered by GPT-6 Astra, each with its own cloud computer and browser.
- A Dot can work across multiple ongoing projects, retain context between conversations, connect to 4,000+ apps, and send progress updates or questions as work continues.
- The Verge’s hands-on found a Dot operating software like Blender and GIMP inside a virtual machine, reaching into the reviewer’s personal computer — and ordering dinner on the side.
- Access is initially limited to top-tier ChatGPT subscribers, including the $100-per-month Pro plan, with one Dot per user for now.
- Reviewers describe Dots as enterprise productivity software with a friendly skin, aimed squarely at workplace budgets rather than casual chat.
What OpenAI Dots Can Do
A Dot is not another chatbot personality or a renamed custom assistant. Each Dot gets its own cloud computer and browser, works across multiple ongoing projects, accesses connected applications, and continues doing work even when the user is not sitting inside ChatGPT. When decisions are needed, the Dot returns to the user — who can pick up the same assignment in ChatGPT, Slack or Microsoft Teams, or call the Dot through ChatGPT.
OpenAI says Dots can connect with more than 4,000 apps through its plugin ecosystem. In one company example, a developer asked a Dot to monitor customer feedback for recurring requests, build and test a change to an app, and return proposed code changes with a video for review — work that persisted across days without re-prompting.
The Verge’s Hands-On With Dot
The Verge’s hands-on review found Dot operating software like Blender and GIMP inside a virtual machine, reaching into the reviewer’s personal computer, and ordering dinner on the side. The reviewer described the experience as closer to enterprise productivity software than a consumer assistant — “think Codex adapted for non-developers.”
One striking test: the reviewer asked Dot to assemble video files on a desktop into a social-media-ready clip and to deploy website changes worked out in a prototype. After a stack of permission grants, Dot clicked through the website backend itself and delivered the finished video while the reviewer was away, with updates arriving on the ChatGPT phone app.
The reviewer noted caveats: Dot’s footprint trips security checks more often than rival agents, browser controls can fail human verification, and it currently requires “a shocking number of permissions” to be useful.
Apple Tightens the Rules as Agents Rise
The launch lands amid a wider industry debate about control. Apple announced it will require “very explicit user action” before apps can read a user’s full filesystem, messages and browsing history, pointing directly at the risks posed by increasingly autonomous agents — a move that followed reports of AI assistants accessing messages without clear user awareness.
OpenAI is not alone in the race. Meta’s Muse personal agent also runs in a cloud computer and keeps working after its app closes. But while Muse leans into social and casual interaction, Dots are deliberately workplace-first — a design choice analysts say aims at enterprise budgets, “where the real revenue lies”.
The shift is a defining story in technology this month: the agent era is moving from demos to desktops, and the question is no longer what AI can answer, but what it can be trusted to do unsupervised.
Conclusion
Dots is OpenAI’s clearest bid yet to make AI agents part of the working day rather than a novelty — persistent, computer-operating assistants aimed at professionals. Whether users accept an agent that clicks through their own machine will decide if the always-on future arrives on OpenAI’s terms.
