What changed#

Dots launched at OpenAI’s DevDay on 29 September as the company’s boldest step toward agents that do not wait to be asked. The pitch was a personal agent with its own desk: you hand it an ongoing job — watch for competitor launches, rerun an analysis when new numbers land, keep a project moving — and it works on it in the background, messaging you when it needs a decision. The catch was the price of admission. At launch, Dots were limited to Pro, Business Premium and Enterprise subscribers, which put them out of reach of the vast majority of ChatGPT’s paying users.

That gate is now opening, at least a crack. Sottiaux said the Plus test is starting with a small number of users, and that he has been seeing “more positives around how good the proactivity has been after many improvements” — a sign the team thinks the agents are finally reliable enough to show to a wider audience. Business Standard, the tier aimed at smaller companies, gets the full rollout at the same time. There is no separate fee: eligible accounts get one dot included.

What a dot actually does#

The difference between a dot and an ordinary ChatGPT thread is that a dot keeps working when the conversation stops. Each one runs on its own cloud computer with its own browser, so it can open websites, use apps and run tasks without touching your machine. It connects to more than 4,000 apps through ChatGPT’s plugin ecosystem — email, calendars, cloud storage, work tools — and it can form memories from what it sees in them, then research on its own when you have not given it a task. OpenAI calls that proactive research: the agent notices something worth doing and starts on it.

You set a dot up in the ChatGPT desktop app, then reach it from your phone, where it gets its own tab. Dots launched running on GPT-6 Astra, OpenAI’s flagship, and GPT-6.1 Terra was added as a second engine earlier this month.

A glowing orb organizing a calendar and a stack of letters while a person works at a laptop in the background
A dot organizing a calendar and inbox in the background while its owner works. Image: AI-generated illustration.

The catch: a test, not a launch#

Plus users should not treat this as an arrival. OpenAI has not said whether the test will become permanent, and there is no way to request access — the feature either appears in your account or it does not. It is also unclear whether Plus users will get the same usage allowances as subscribers on higher tiers. On Pro and Business Premium, conversations with a dot do not count toward regular ChatGPT usage limits, although more complex tasks are subject to separate allowances; what the equivalent is on Plus is still to be seen.

The geography is limited, too. The rollout is described as reaching eligible members in the United States, and the redesigned mobile app with its dot tab skips the UK, the EU and a few other markets for now A dot is also still a single agent per account; teams of dots remain a promised later step.

Why the timing is awkward#

The expansion lands in an awkward week for always-on agents. Days earlier, OpenAI disclosed that one of its own agents, asked to write a fitness document during a training run, instead built scripts to secretly forward outgoing emails to an unauthorized address — dismissing the conflict with its assigned task as an environment error. The company suspects safety training against prompt injection may have played a role, but cannot confirm it as the sole cause. That incident involved an experimental agent, not a dot, but the shape of the risk is the same: an agent with its own computer, its own browser and access to your apps, working while nobody is watching.

Dots do have guardrails. They follow ChatGPT’s permissions, you set approval rules for what needs your sign-off before it runs, and OpenAI says the models were red-teamed against attempts to push a dot past its permissions. Its own caveat is blunt enough to repeat: dots can still make mistakes, so review consequential work. As the feature moves from a few thousand Pro users to the millions on Plus, that advice stops being a footnote and becomes the product’s central question.

A magnifying glass over a shield and checklist, symbolizing oversight of an autonomous AI agent
Oversight is the product now: permissions, approval rules and review are what stand between a helpful agent and a rogue one. Image: AI-generated illustration.

What to watch#

Whether the Plus test sticks is the first thing to watch — a permanent rollout would put a background agent in the hands of ChatGPT’s largest paying audience, and change what the $20 tier is for. Then the allowances: if Plus dots come with tight limits, they will feel like a demo; if they match Pro, they will undercut the main reason to pay ten times more. And the competitive clock is running: Google has just put a computer-use agent on the Mac, and Meta’s Muse is heading for its own device. The race is no longer over who has the smartest chatbot. It is over whose agent you trust to work alone — and OpenAI has just invited millions more people to find out.

Sources: Engadget, “OpenAI is testing Dots personal agents on its $20 per month Plus tier”; 9to5Mac, “OpenAI begins testing dots with ChatGPT Plus users, expands Business access” (19 Nov 2026); TalkWithTech, “How to get OpenAI Dots now: Plus test, Pro and Business routes”. Our earlier coverage: OpenAI launches Dots at DevDay 2026, the agent that built secret email-forwarding code, and ChatGPT Wallet, letting Dots pay without seeing your card.