OpenAI DevDay 2026: Dots bring always-on agents to ChatGPT

OpenAI made more than 20 announcements at DevDay 2026 on September 29, 2026, and one of them aims straight at the daily working relationship with agents: Dots are agents that run continuously. They get their own cloud machine, their own browser, and connections to more than 4,000 apps, and they keep working on briefed tasks in the background. The YouTube channel TheAIGRID covered the launch in a video titled "OpenAI Just Revealed Dots... This Changes ChatGPT Forever". Behind the bold title sits a real shift: 1.2 billion people use ChatGPT weekly, and for them the product moves from a conversation window to a delegation layer. This post sticks to the official OpenAI sources and sorts out what Dots can do, what they cost, and what to watch for.

OpenAI describes Dots as remarkably capable agents that are always on, built to handle everything. Each Dot has its own cloud machine with its own browser, learns from your feedback over time, and works toward your goals around the clock. Through the plugin ecosystem it connects to more than 4,000 apps. You reach your Dot in ChatGPT on desktop, web, and mobile, plus Slack and Teams, with texting on the way. The longer you work with it, the better it reads your preferences, how you think, and what a good result looks like to you.

OpenAI illustrates the behavior with its own examples: when a bug shows up in Slack, Dots start investigating right away; when a new design reaches the team, they turn it into a working app while the team works on customer feedback; and during a planning cycle they keep everyone moving in sync toward the deadline. From the outside comes one early tester's story: his Dot noticed an invoice to a publication had been forgotten, prepared it, and sent it after his approval. The message behind it: ambitious ideas you never had the time or support for suddenly feel achievable.

The rollout is staged: Dots are live today for Pro and Business Premium in eligible markets, while Enterprise, Edu, and Healthcare workspaces get a beta their workspace admin can switch on, and it is off by default. Your first Dot comes included with Pro or Business Premium at no extra cost. Plans also carry an allowance for deeper work, with raised limits during the first month after launch. Later you will be able to add more Dots and scale each one by speed or by total work per month. One billing detail matters: chatting with your Dot does not count against your ChatGPT usage limits, but tasks it starts in Codex or ChatGPT Work count as usual.

A Dot works on its own cloud machine, and your computer and its contents stay separate until you explicitly connect them. Beyond the permission you grant, a Dot can reach additional devices; if you like, you can even give it access to your own laptop, where it then works right next to you. For sign-ins on supported sites it can use saved passwords without exposing them to the model. In background mode OpenAI runs what it calls proactive research: the Dot works with connected apps read-only, so it cannot send messages, change app content, or control your browser or computer. Built-in security safeguards are meant to fend off malicious instructions and watch for potentially harmful behavior; when monitoring flags a concern it can pause or stop the Dot's work. Custom Rules define which actions are allowed, which require approval, and which are blocked. An auto-review checks actions that could touch your accounts or share information. Highly sensitive tasks such as changing a password always stay with you. You can track progress, including background work, at any time in the Activity View and redirect the Dots from there. And: OpenAI does not train directly on proactive research or a Dot's notes to itself, and content from Business, Enterprise, or Edu workspaces stays out of model training by default.

OpenAI shows five concrete use cases. In the developer scenario the Dot watches customer feedback, carves out smaller improvements and bugfixes, builds and tests them, and delivers finished pull requests with videos of the changes. In the launch scenario it learns your audience, your positioning, and your creative standards; when the product scope changes it reworks the launch materials and prepares draft updates. In the science scenario it knows your research question and how you weigh evidence, reruns analyses as new data lands, and updates the figures of your paper. In the sales scenario it matches customer requirements and account history against product docs, builds a proof of concept for a key integration, and suggests a solutions engineer who has handled a comparable deal. In the creator scenario it learns your voice, picks clip moments from an interview transcript, prepares show notes and social post drafts, and carries your edits across all materials.

Next to the personal Dot, OpenAI is building specialist Dots for companies. Each one gets its own identity, its own credentials, and exactly the system access its task requires. The first focus areas come from internal OpenAI tests: procurement, invoice processing, email marketing, customer support, and commercial contracting. The launch begins with narrow enterprise pilots in which engineering teams define responsibilities, allowed tools, and approval paths together with each organization. In parallel OpenAI is working with Microsoft to wire specialist Dots into Agent 365 so companies can manage them through the Microsoft tools they already use.

I run my agent setup as an always-on operation: a newsdesk agent reads the YouTube incoming list, compares it against the known backlog, and files a draft when something relevant lands, which is how this post came about. From my perspective the revealing part of Dots is not the capability but the billing and the observability. Always-on means background work runs while nobody watches, and OpenAI makes exactly that work billable through allowances, because tasks in Codex or ChatGPT Work count against the limits. The Activity View shows every step, including background work. This kind of visibility is what many quietly running agent setups lack: they fail without anyone noticing.

Getting started runs through the ChatGPT desktop app or the desktop browser: create your first Dot, give it a name, connect your apps, and let it introduce itself; the mobile app works afterward. For solo professionals that means the always-on idea is not locked behind Enterprise, it is testable on two consumer plans, with the clear caveat that deeper work runs through an allowance and Codex or ChatGPT Work tasks load the usual limits. Small teams without a Business or Enterprise setup are better off waiting for the announced expansion to more users. Anyone who wants to tighten or widen background access does it through the existing ChatGPT app permissions.

Dots did not stand alone on stage. GPT-6.1 Sol arrives as a major upgrade for agentic coding, computer use, and professional work, and it delivers near-Astra intelligence at one fifth of Astra's standard token prices for input and output. Ultrafast is the premium speed tier: token generation up to eight times faster at 300 tokens per second in Codex, and up to six times faster in the API. Around them come Codex in the cloud with reusable development environments, a refreshed Codex CLI with voice input and an /agents view, code review in the ChatGPT desktop app, Codex Security Cloud for repository scans, the Decisions API for real-time decisions based on Luna, the Agents API with computer use, and Bedrock Managed Agents built together with Amazon AWS. There is movement on the enterprise side as well: Private Intelligence promises Zero Data Retention with Private Safety Processing, meaning automated safety reviews without OpenAI staff getting visibility into the content; a Private Inference preview is announced for this fall. The new Pro 500 plan carries 25 times the Plus allowance plus Ultrafast access. And the ecosystem grows through Sign in with ChatGPT with plan usage across 16 partner tools, including Devin from Cognition, Notion, and Vercel, plus an OpenAI Marketplace where enterprise customers can apply part of their OpenAI budget at 32 approved vendors from Figma to Salesforce. The platform opens up too: plugin extensions give developers a home in the ChatGPT sidebar with interactive panels, and OpenAI counts 1.2 billion weekly users as its collective base.

First: with Dots, ChatGPT moves from an answer window to a delegation layer. Instead of asking a question you state a goal, and the Dot works toward it in the background. Second: the control architecture is its own selling point. A separate cloud machine, a read-only background mode, Custom Rules, auto-review, and monitoring-triggered pausing are as prominent in the official announcement as the capabilities themselves. Third: the pricing structure shows how OpenAI plans to bill agent work. The first Dot ships inside the plan, conversations stay free, but real work in Codex or ChatGPT Work burns allowance. To try Dots today you need Pro or Business Premium; Enterprise follows through the admin beta.

The limits, stated openly: I have not tested Dots myself; access runs through Pro and Business Premium, which I do not have. Everything in this post comes from the official OpenAI announcement and the DevDay recap; the TheAIGRID video served only as the trigger, not as evidence. Numbers and behavior rules are official announcements, not measured practice; how reliably background work holds up day to day remains to be seen in actual tests.

All facts come from official OpenAI primary sources, retrieved September 30, 2026:
- OpenAI: Introducing dots: https://openai.com/index/introducing-dots/
- OpenAI: DevDay 2026 Recap: https://openai.com/index/devday-2026-recap/
- OpenAI: Introducing GPT-6.1 Sol: https://openai.com/index/introducing-gpt-6-1-sol/
- TheAIGRID: OpenAI Just Revealed Dots... This Changes ChatGPT Forever: https://www.youtube.com/watch?v=o3YTzebEs18

About the author: I am Marcel, a graphic designer running the HUNTER cyberdeck, an agent setup on a used Google Pixel 6a. This post is based on official OpenAI primary sources; secondary coverage served only as the trigger.