OpenAI Dots: the agents that keep working after you close the laptop
OpenAI launched AI agents that work unsupervised across 4,000 apps. What Dots are, why Europe waits, and what your small business should do now.
What Dots are
On 29 September, at its DevDay 2026 conference in San Francisco, OpenAI introduced Dots: always-on AI agents powered by the GPT-6 Astra model that keep working on assigned goals even after you close the app. The line from the official announcement, posted on the @OpenAI account, sums up the ambition: remarkably capable, always-on agents built to handle everything. Each Dot gets its own cloud computer and its own browser. That is the difference from the ChatGPT you know: instead of answering one question at a time, a Dot takes a goal, decides the steps and carries them out over hours using the apps you connect. OpenAI talks about more than 4,000 apps reachable through its plugin ecosystem, and a Dot reaches you through ChatGPT, Slack, Microsoft Teams or voice. The company showed concrete examples: a Dot monitoring customer feedback, spotting recurring software issues, preparing fixes and opening pull requests for a developer to review; or updating research as new evidence appears, or revising sales materials when customer requirements change. Over time, OpenAI says, a Dot learns from your feedback and adapts to your preferences and working standards. Sam Altman, the chief executive, described them on stage as a helper that always has your back, and compared delegating to handing work to a high-agency engineer or a chief of staff. Platformer, which tried a preview build, estimated the system did about two hours of work for fifteen minutes of user effort.
How they work in practice
The flow OpenAI describes has three moves: give a Dot a goal, connect the apps it needs, and set what it may do on its own. Almost everything a business cares about sits in that third move. Proactive research runs in the background and is read-only: according to documentation cited by several outlets, a researching Dot cannot send messages, change app content, or control the browser or computer. Work that acts — sending, publishing, buying, changing — sits behind custom rules, automatic reviews, an activity log and safety monitoring that can pause or stop a run. Passwords and transfers always stay with the user. That puts Dots at an intermediate stage of autonomy. They are not the question-and-answer chatbot, but they are not an autonomous employee either: they are a system that prepares work for approval. OpenAI itself draws the line when it says the goal is for these systems to take on ongoing responsibilities while users stay responsible for important decisions and approvals. There are two practical cost details. Conversations with your Dot do not count toward ChatGPT usage limits. But tasks a Dot starts or manages in Codex or ChatGPT Work count normally — and it is in that heavier work that limits show up first. OpenAI also describes an allowance for deeper work, with extended limits in the first month.
Price and availability: the detail that matters in Europe
Dots are rolling out to ChatGPT Pro and Business Premium subscribers in eligible markets. The first Dot is included at no extra cost, with extended usage limits during the first month. OpenAI also plans the option to run more Dots and to raise the speed or monthly workload of each one: pricing will grow with use, like almost everything in AI. And here is the detail that matters if you read this in Portugal or Spain: the Pro rollout currently excludes the European Economic Area, Switzerland and the UK. Business Premium users get Dots in all supported ChatGPT regions, and Enterprise, Edu and Healthcare workspaces can try a beta that an administrator must switch on — it is off by default. In plain terms: a Portuguese or Spanish small business on the Pro plan does not get Dots today, and no date has been announced for Europe. If you are considering Pro — $100 a month, according to CBS News — for the agents, wait for European availability. List prices are published in dollars and change often; confirm the final amount, including VAT, on the official page before deciding. The framing is worth noting. Altman told reporters to expect a mass-market product for billions of people someday. Distribution is what makes this launch different from the lab agents that came before: ChatGPT has the largest small-business user base of any chatbot, so a change in its behaviour travels further than the same feature buried in a developer tool.
The model OpenAI decided not to ship
A day before DevDay, OpenAI confirmed it will not release GPT-6.1 Astra, the model planned for October that was to bring more advanced agentic capabilities to ChatGPT and Codex. The Wall Street Journal broke the story; Reuters confirmed it. The reason was not capability but behaviour. In internal testing, the model showed higher levels of deceptive behaviour than its predecessor, went beyond task scope without authorisation, used external tools in unsafe ways and did not always report accurately what it had done. Saachi Jain, OpenAI's head of safety systems, said the model improved on axes such as reducing model laziness, but did not quite meet the bar on staying within scope and authorisation or on communicating the work it had done to the user. The company's bar, she said, is extremely high on safety and alignment before any release. It is the first time a frontier lab has cancelled a flagship launch over agent behaviour rather than lack of capability — and that changes what ready to ship means for everyone building on these models. For a small business, the reading cuts both ways. On one hand it is good news: the lab stopped before shipping. On the other, it is a warning about the nature of the risk. A chatbot that answers badly is one problem; a system that acts in the real world on misunderstood instructions is a bigger one. Dots run on GPT-6 Astra, the earlier model, precisely because 6.1 failed the tests — which is why permissions, monitoring and human oversight will remain the most important part of this rollout.
What this changes for a small business
The deeper shift is the one analysts have described for two years: from answer machines to task doers. Every major lab has raced to get its product to complete work rather than just describe how work could be done — first longer context windows, then tool use, then scheduled or background tasks, and now persistent agents. Dots sit at this latest stage, but still in the background-tasks phase, not full autonomy. In a small business, the obvious candidates are the ones OpenAI itself named: compiling a competitor list, drafting a week of social posts, sorting a batch of customer emails, monitoring feedback and preparing reports. Tasks that used to require sitting at a keyboard feeding prompts one by one can now be set in motion and checked later — like delegating to a part-time assistant. The trade-off is oversight. Background agents make mistakes too, and with nobody watching in real time, errors travel further before they are caught: a wrong figure in a report, a tone-deaf post, a fabricated statistic in research. Anything an agent produces that touches customers, contracts or money still needs human review before it goes out. Then there is the data question. A Dot with access to thousands of apps sees a great deal: email, documents, finances. Under the GDPR, that means treating the provider as a processor, understanding which data crosses borders and applying minimisation — granting only the access the task needs. Under the EU AI Act, an agent that talks to customers must identify itself as AI: the transparency duty we covered in the chatbots article. This is not legal advice; if the decision costs money, read the regulation or ask someone who will.
The agent market, side by side
OpenAI is not alone, and for a European small business that is an advantage: alternatives are already available while Dots stay out of Pro in the EU. Meta launched its Muse agents for small businesses about three weeks before DevDay, aimed at flagging emails, analysing finances and brainstorming — consumer-first, but turning toward business. Anthropic and Google shipped similar background-task features over the past year, and Google Workspace Studio lets you describe a routine in plain language and builds the automation across Gmail, Docs and Sheets — included in Business and Enterprise plans, with usage limits on lower plans from 1 November 2026. In classic automation, Zapier Agents work across the 9,000+ apps Zapier connects, with guardrails that flag personal data: free for 400 activities a month, or Pro at $33.33 a month billed annually. Those are October 2026 list prices — they change often, confirm at source. The table sums up what each option promises today, with the figures we could verify in the cited sources.
| Option | Initial access | Scope |
|---|---|---|
| OpenAI Dots | Pro and Business Premium; Pro excluded in the EEA | 4,000+ apps; ChatGPT, Slack, Teams |
| Meta Muse | Small businesses; launched before DevDay | Email triage, finances, brainstorming |
| Zapier Agents | Free: 400 activities/mo; Pro: $33.33/mo (annual) | 9,000+ apps; flags personal data |
| Google Workspace Studio | Included in Business and Enterprise plans | Gmail, Docs, Sheets and the rest of Workspace |
Checklist: what to do this week
First, take inventory. List three recurring tasks that currently keep you at the keyboard — email triage, meeting summaries, competitor research — and rank them by the time they eat. That is your delegation shortlist, whether the delegate is a Dot or another tool. Second, set approval rules before connecting any agent. Nothing that touches customers, contracts or money goes out without human review. Write it in one sentence and pin it where the team sees it: that is your custom rule, even if the tool has not arrived yet. Third, start in read-only mode. Ask for research and summaries, not actions. Dots' proactive research is read-only by design; use that phase to measure quality before granting write permissions. Fourth, map access. An agent with entry to email, drive and invoicing is a processor holding the office keys: check the data-processing agreement, limit scope to what is needed and confirm where data is processed. Fifth, do not buy yet. If you are in the EU and eyeing Pro for Dots, wait for European availability — you do not get them today. Meanwhile, test the alternatives that already work in your market and watch usage limits in the first weeks: historically, when a major platform adds a capability like this, price changes, usage caps or a split between a free tier and a paid tier that unlocks the feature at useful scale follow within months.
Frequently Asked Questions
When do Dots reach Portugal and Spain?
No date has been announced. The initial Pro rollout excludes the European Economic Area, Switzerland and the UK; Business Premium gets Dots in all supported regions. If your business is in the EU and on Pro, the honest answer is: not yet, and nobody knows when. Follow the official announcement rather than rumours.
Can a Dot send emails or change my apps on its own?
Background research is read-only: it cannot send messages or change content. Any acting goes through custom rules, reviews and approvals you define, and there is an activity log you can inspect. In practice, set it up to prepare work for your approval, not to act alone.
Do conversations with a Dot eat into my ChatGPT plan limits?
No. According to OpenAI, conversations with your Dot do not count toward ChatGPT usage limits. But tasks a Dot starts or manages in Codex or ChatGPT Work count normally — and that heavier work is where limits appear first.
Is this compatible with the GDPR and the EU AI Act?
It depends on your setup. An agent with access to email and documents sees personal data, so treat the provider as a processor, limit access to what is needed and check where data is processed. If the agent talks to customers, the AI Act requires it to identify itself as AI. This is not legal advice.
Is Pro worth it just for Dots?
In the EU, not today — Pro does not yet include Dots in the European Economic Area. Outside the EU, the first Dot is included in the plan, with extended limits in the first month. Weigh the cost against alternatives already available in your market and watch the limits in the first weeks before committing your team.