OpenAI's dots are always-on AI agents inside ChatGPT: each one has its own cloud computer, connects to your apps and keeps moving your work forward even when you're away from the screen. You talk to them in ChatGPT, Slack, Teams or by text message, and they ask for your approval before anything that matters.
OpenAI introduced them on September 29, 2026, the same day as its DevDay and the launch of GPT-6.1 Sol. It's the clearest sign of how fast things are moving: what was a custom agent project a few months ago now comes included in a ChatGPT plan.
What are OpenAI dots?
According to OpenAI's official announcement, a dot is an agent powered by GPT-6 Astra that is onboarded "with almost everything you'd give a new employee":
- Its own identity to manage access and permissions.
- A cloud computer with its own browser; it can write and test code when needed.
- Connections to more than 4,000 apps through OpenAI's plugin ecosystem.
- Memory of your preferences: it learns from your feedback and from what you consider good work.
- 24/7 availability and the ability to run several projects at once without a separate conversation for each.
The difference from a regular chat is that a dot doesn't wait for you to ask. It keeps track of what's in progress and runs "proactive research" in the background, with read-only tools, to get ahead of what you'll need.
A day with a dot: autonomy and working from anywhere
The launch video tells the story better than any spec sheet. A person creates a dot, names it "Alfred" and gets on with the day. Meanwhile, the dot:
- 1Gets the website ready for the fall launch using the approved designs.
- 2Takes the customer experience numbers that came in on Teams and updates the board meeting deck.
- 3Spots a pending app migration in the Google Meet notes and maps out the changes.
- 4Handles a personal surprise: the wedding cake vendor canceled, and the dot already has a backup and a tasting booked.
Later, the person checks in from wherever they are: asks to swap two photos and the dot publishes the site; reorders the deck to lead with customer experience; the migration tests come back with no regressions, and the dot shares the results on Slack and gets the rollout ready for approval. At the end, it moves a finance review that clashed with a daughter's recital.
The dot executes; the person decides. That's the new way of working OpenAI is proposing.
What the video shows isn't a faster chatbot. It's delegating complete work, with follow-through, from whatever channel is at hand, while keeping control at the moments that count.
What it looks like in a company: a day with Samantha
In our video we take the same idea into a company. Samantha is the agent; Miso runs sales and Ryu is the owner. Neither of them sits in front of a screen waiting: the work moves forward on its own and they decide from wherever they are.
In the morning, Samantha tells Miso that fourteen quote requests came in overnight and the proposals are ready with this week's prices, and that the payment reminders already went out. Miso just reviews and says "send them". Later, Ryu is working from his living room when Samantha lets him know a customer approved the proposal: he asks to enter the order and schedule the kickoff, and billing already has it.
And the payoff is the same as in OpenAI's video: time given back. Samantha notices the finance meeting clashed with Miso's daughter's school festival, moves it and confirms with everyone while she's at the park.
What would a dot look like in a company?
The announcement's own examples (a forgotten invoice, a proposal that changes with the customer's requirements, a report that updates with new data) map directly onto a mid-sized company's operations:
| Today | With an agent like a dot |
|---|---|
| The sales lead rewrites the proposal every time the customer changes a requirement. | The agent compares the new requirements with the proposal, flags what's missing and prepares the updated version for approval. |
| Someone remembers late that a service was never invoiced. | The agent spots it, prepares the invoice and sends it once you approve. |
| The leadership report is assembled by hand first thing Monday. | The agent gathers each team's numbers and has the report ready with what changed. |
| Meeting decisions get lost in the notes. | The agent turns decisions into tasks, follows up and flags anything that stalls. |
| Two meetings collide and someone spends the morning moving calendars. | The agent proposes a new time, coordinates it and prepares your notes. |
In every case the pattern is the one from the video: work moves forward on its own and a person approves what has consequences.
What about control? Always with your approval
The safety design is what makes this workable in a company. According to OpenAI:
- Each dot works on its own cloud computer; your device stays separate unless you choose to connect it.
- Proactive research uses read-only tools: it doesn't send messages, change apps or control your computer.
- Built-in rules define when to act independently and when to ask, and you can add custom rules to allow, require approval for or block actions.
- Automatic review checks actions that affect your accounts or share information; sensitive tasks like changing a password always stay with you.
- In Business and Enterprise, OpenAI doesn't use your workspace content to train its models by default.
Who can use dots and what do they cost?
Rollout is gradual. According to OpenAI's help center, dots are reaching ChatGPT Pro users outside the European Economic Area, Switzerland and the UK, and Business Premium users in every region where ChatGPT is available. In Enterprise, Edu and Healthcare they're in beta and an admin has to enable them.
- Your first dot is included in Pro and Business Premium at no extra cost, with a usage allowance for deeper work.
- Talking to your dot doesn't count toward your ChatGPT limits; tasks it starts in Codex or ChatGPT Work do.
- For companies, OpenAI started pilots of "specialized dots" (procurement, invoice processing, customer support, email marketing, sales contracting), set up with its engineering team, and is working with Microsoft to manage them from Agent 365.
The question is no longer whether to use AI
Plenty of companies are still deciding whether to use artificial intelligence. Meanwhile, in a single week OpenAI shipped a new model, a full DevDay of developer tools and agents that work on their own. The "should we use AI" debate is already behind us.
The real question is how long it will take you to start using what already exists, and what arriving late will cost you.
The edge doesn't go to whoever waits for the technology to mature. It goes to whoever already knows which processes to delegate, with what data and under what rules, so they can use each new tool the day it ships.
What a dot doesn't solve on its own in a company
A dot is a great assistant for one person. For an agent to run a company process, you usually need pieces that don't come in the box:
- Systems without a plugin: the ERP, the in-house system or the database where the real operation lives.
- WhatsApp as a customer channel: it isn't among the channels announced for dots (ChatGPT, Slack, Teams and text messages). If your operation runs on it, it's integrated separately; see our guide to WhatsApp ERP and CRM integration.
- Rules by team: who can approve a discount, which data each role sees, which actions are never automated.
- Data that can't leave the company: for sensitive information, an agent running on local, private AI is sometimes the better fit.
- Measurement: knowing how much time, error rate or revenue the agent changed, to decide what's next.
How WITS helps you put it to work and grow your business
At WITS, a software and AI team in Guadalajara, Mexico, we build AI agents for businesses: we take the same idea from the video (autonomy, working from any channel and human approval) and bring it into your company's processes.
- 1We pick the processes worth delegating first: repetitive, costly in hours and with clear rules.
- 2We prepare data, access and permissions so the agent works with reliable information.
- 3We connect what a dot can't connect on its own: ERP, CRM, WhatsApp and in-house systems, with process automation.
- 4We define the approval rules: what it does alone, what it asks you and what it never touches.
- 5If a dot is enough, we help you set it up and get the most out of it; if not, we build a custom agent.
- 6We measure the result and extend to the next processes.
If you want to see what this would look like in your operation, start with AI for business or how we work with US companies.
