OpenAI’s Dots: The Always-On AI Assistant Built to Handle Everything

https://bitbiased.ai/ai-automation-services OpenAI just gave ChatGPT its own computer — an always-on cloud machine that can work on your goals even when you’re not actively chatting with it. Announced at DevDay on September 29, 2026, OpenAI calls it a “Dot.” Instead of behaving like another ChatGPT session, a Dot is designed to hold ongoing responsibilities: monitoring connected apps, working across sessions, running code, browsing websites, and proactively messaging you when something needs attention. But the bigger story is what OpenAI deliberately won’t let it do without you. Each Dot runs on GPT-6 Astra and gets a dedicated cloud virtual machine with a Chromium browser, filesystem, and access to Codex for coding tasks. OpenAI says it can connect to more than 4,000 apps through the existing ChatGPT integration ecosystem, including Gmail, Google Calendar, Slack, GitHub, Notion, and Salesforce. The persistence is what makes this different. Files can remain on the Dot’s machine, connected login sessions carry over, and recurring tasks can keep running without you repeatedly issuing the same prompt. The Dot can also carry context across supported surfaces such as ChatGPT, Slack, and Microsoft Teams. OpenAI demonstrated what that looks like with several scenarios. One Dot noticed its user working late, found dinner options, and proactively messaged them. Another detected a mobile checkout bug, accessed the connected GitHub repository, wrote a fix, ran tests, and opened a pull request. Another identified a scheduling conflict and prepared a reschedule. A research example showed a Dot automatically rerunning an analysis whenever new experimental data arrived. These were curated OpenAI demonstrations, though — not independent stress tests. And that's where the permission system becomes important. OpenAI says every planned action passes through an auto-review layer. A Dot may be allowed to handle lower-risk work within its granted permissions, but sensitive actions such as purchases, password changes, and software installation remain with the user. Custom Rules can further define exactly what an individual Dot is allowed or forbidden to do. That architecture also matters for prompt injection. A persistent agent reading webpages and emails creates a different security problem from a chatbot answering a single prompt. OpenAI says background reading uses restricted tools and planned actions still pass through its review system, but how that protection performs against real-world attacks remains something independent testing will need to establish. There’s another problem: persistent mistakes. A bad ChatGPT answer might disappear into an old conversation. A bad assumption made by an agent working independently for several days could influence multiple later actions before a human notices. OpenAI acknowledges Dots can make mistakes and says they can stop and ask for help when authentication expires, a website fails, or safety monitoring intervenes. Not everything underneath Dots is completely new, either. Browser automation builds on ideas already seen in Operator, coding uses Codex, and thousands of integrations already existed. The major change is packaging those capabilities around a persistent identity, dedicated cloud computer, recurring responsibilities, and an action-review system. OpenAI has also discussed a future involving teams of specialized Dots coordinating with each other. That isn't the product shipping today. The launch has significant limitations: one Dot per account, ChatGPT Pro or Business Premium access in eligible markets, no public API, unspecified compute and memory limits, and Codex-heavy workloads still counting against normal usage limits. The script cites ChatGPT Pro at $100 per month. So is this really “AI that never stops”? Dots can act between conversations without waiting for another prompt, which is a meaningful change. But they're still operating inside boundaries established by a human, with sensitive actions requiring approval. The interesting question isn't whether OpenAI has created a fully autonomous digital employee. It's what happens when an AI assistant stops being something you repeatedly ask for help — and starts holding a responsibility over time. CHAPTERS 00:00 OpenAI Gave ChatGPT Its Own Computer 01:39 Not Another ChatGPT Tab 03:17 Its Own Cloud Computer 04:53 Four Things OpenAI Actually Showed It Doing 06:15 Everything That Still Needs Your Permission 07:35 What Happens When It Gets It Wrong 08:30 How Much of This Is Actually New 09:42 The Fine Print 11:16 Is This Actually AI That Never Stops? #OpenAI #ChatGPT #GPT6Astra #AIAgents #ArtificialIntelligence