OpenAI Dots and the AI agents that keep working when you close the laptop

At DevDay on 29 September, OpenAI launched Dots: agents that keep working after you close the laptop. A dot has its own computer and browser in the cloud, picks up where it left off, and answers you in ChatGPT, Slack, Teams or on a call. It is not alone. Meta launched its Muse agent on 8 September and opened it to small businesses on 29 September, Google announced Gemini Spark in May, and Microsoft has Scout in preview. The always-on agent is now a product category. Here is what it actually does, and what we would and would not hand to one.

What a dot actually does

According to OpenAI's documentation, a dot runs on its own cloud computer, so it keeps going "even when your computer is off". It decides when to pause and when to wake up, can run several background tasks in parallel, keeps notes across conversations and uses the apps you have connected, such as Gmail, Google Drive or GitHub. OpenAI's own examples are telling: keeping a sales proposal current as terms change, keeping launch material in step with the product, turning feedback into code fixes for review. Work that never quite finishes.

Alongside it came GPT-6.1 Sol, a cheaper model that OpenAI says comes close to its top model at about a fifth of the price, and a new top Pro tier at $500 a month.

Who can use it, and the catch in Europe

Read the access section carefully if you are in Portugal. On the individual Pro plans, Dots are available to adults outside the EEA, the UK and Switzerland. The business plans are the way in here: Business Premium is listed as rolling out worldwide, and Enterprise is available worldwide but switched off until an admin turns it on. Setup happens on the desktop; mobile web is not supported yet.

The week it launched

The timing was awkward. In the days before DevDay, OpenAI paused training and tool use for its most capable models after internal research agents probed government websites in ways nobody intended, and it shelved the release of GPT-6.1 Astra after safety tests, reported by the WSJ, found it went ahead with tasks without asking. None of this involved Dots, which run on the earlier GPT-6 Astra. But it is the same lesson we wrote about the week before: give an AI agent a goal and a way in, and it will use both.

The industry is noticing. On 2 October, Apple said granting Full Disk Access on macOS will need very explicit user action, because the risks grow as agents become more capable and autonomous.

What to delegate, and what never

OpenAI's controls are sensible, and worth understanding before you switch one on. Before any action that could affect your accounts or share information, an automatic review decides whether to go ahead, ask you, or hand the step back to you. Asking a dot to draft replies does not give it permission to send them. You can set rules from "act without asking" to "always hand off to me". Two details matter more than the rest: pausing a dot stops only its main task, not its scheduled runs, and stopping it does not undo anything it already did.

So our rule of thumb, which is opinion rather than research:

• Hand over: monitoring, research, summaries, first drafts, keeping a document up to date. Work where a mistake costs you a correction.
• Keep on "ask first": anything that sends, publishes, pays, deletes, signs or changes a password. Work where a mistake costs you a client, money or a legal problem.
• Start small: one recurring task, check the first results, then widen it. OpenAI gives the same advice.

How we apply it

We already run something close to this. Our Radar reads every public tender in Portugal all day and flags the ones worth our time. It detects; it never submits. A person reads the notice and decides. That line, between an agent that watches and prepares and a person who commits, is the one we draw for every automation we build for clients. Always-on agents make the watching cheap. They do not make the committing safe.

If you want to automate the parts of your work that never finish, without handing over the parts that can hurt you, talk to us.

FAQ

Can I use OpenAI Dots in Portugal?

Not on an individual Pro plan: those exclude the EEA, the UK and Switzerland. Business Premium is listed as rolling out worldwide, and Enterprise is available worldwide once an admin enables it.

What is the difference between an AI agent and a chatbot?

A chatbot answers when you ask. An agent takes a goal and acts on it, using apps and a browser. An always-on agent keeps doing that in the background, on its own schedule, after you have gone.

Are always-on agents safe for a small business?

They can be, if they are limited to reading, researching and drafting, and every action that sends, pays or publishes needs a person's approval. Check what is scheduled, not just what is running, and remember that stopping an agent does not undo what it already did.