OpenAI Dots: Give Your Business an AI Coworker That Never Sleeps

What OpenAI launched on September 29

At its DevDay event in San Francisco on September 29, 2026, OpenAI introduced Dots — which it describes as "remarkably capable, always-on agents built to handle everything," according to TechCrunch's report of the announcement. Each Dot is an AI agent powered by GPT-6 Astra, and it gets its own cloud computer and web browser. Unlike a Codex task or a normal ChatGPT session, a Dot does not end when you stop typing. It keeps pursuing the goals you set while you get on with your day.

You reach it the way you already talk: inside ChatGPT on web, desktop and mobile, or by messaging it in Slack and Microsoft Teams, with text messaging on the way. It can connect to more than 4,000 apps through ChatGPT's plugin ecosystem, per VentureBeat's report of the launch details, and OpenAI says your first Dot is included with ChatGPT Pro and Business Premium at no extra cost. Enterprise, Edu and Healthcare workspaces get a beta that stays switched off until an administrator enables it.

OpenAI's own examples, reported by VentureBeat from the keynote, hint at the ambition: a developer deploying a Dot to watch incoming customer feedback and implement small fixes, a scientist having a Dot rerun analysis as new data arrives, a sales Dot keeping a proposal current as a deal changes. Sam Altman told the DevDay audience you can "delegate ambitious pieces of work the way you would to a high agency engineer or a chief of staff," per AI Weekly's report. Platformer's Casey Newton, testing a preview build, estimated it did about two hours of work for about fifteen minutes of his effort, also reported by AI Weekly.

How a Dot is different from a chatbot

A chatbot waits. You ask, it answers, the window closes, and the context ends with it. A Dot starts. You give it a standing goal — "watch our incoming quotation requests and draft a follow-up for every quote older than three days" — and it keeps working in the background, checking your connected apps and picking the work back up where it left off.

There are guardrails on the autonomy, and they matter for a business. Background or proactive research is read-only, according to ExplainX's launch breakdown: the Dot can look across your connected apps but cannot send messages, edit content, or take over the browser and computer without you. It asks before sensitive actions. Chatting with your Dot does not count toward your ChatGPT usage limits — although tasks it kicks off in Codex or ChatGPT Work do, per The Web Tier's launch breakdown. That division is worth remembering: talking is cheap, doing is metered.

The mindset shift is the real product. Businesses have spent years learning to write better prompts. Dots move the job one level up. Instead of prompting well once, you describe a job well once — and the agent keeps doing it.

The practical use case: a follow-up desk that never sleeps

Quotations are where small businesses quietly leak money. You send a quote on Monday, the customer goes quiet, and by Friday the thread is buried under forty new emails. Nobody was assigned to chase it, so nobody did. Here is how one Dot takes that job — not as a demo, but as a weekly routine you can set up in an afternoon.

Step 1: Write a one-paragraph job description

Before opening ChatGPT, write down exactly what the Dot owns. One job, not five. For example: "Every weekday morning, scan my email for quotation requests and sent quotes. Flag any quote older than three days with no reply, draft a polite follow-up in my tone, and prepare a one-page summary of pending quotes with amounts and last contact date."

If you cannot write the job in one paragraph, the job is too big. Split it.

Step 2: Connect only the tools the job needs

Create your Dot in the ChatGPT desktop app or a desktop browser, then connect just the apps this job touches — your email and your calendar, maybe your Drive folder of saved quotes. Every connected app is a door; only open the doors this job walks through. You can message the Dot later from the mobile app, but setup belongs on desktop where you can see what you are granting.

Step 3: Give it a starter instruction you could hand a new hire

Paste the job description as the Dot's standing instruction, plus two rules in plain language:

"Rule 1: Never send anything to a customer without my approval — prepare drafts and wait.

Rule 2: Every morning by 9, give me the pending-quotes summary: customer name, quote amount, days since last contact, and one suggested next step."

A good test of the instruction: if you handed the same paragraph to a smart new hire, would they know what to do? If yes, the Dot will too.

Step 4: Set the one approval rule that matters

The difference between a useful Dot and a liability is one sentence: it drafts, you send. Keep that sentence in the instruction for the first month. OpenAI designed Dots to ask before sensitive actions — keep that behaviour switched on and treat the morning summary as your review checkpoint. You are the manager; the Dot is the trainee who happens to never sleep.

Step 5: Run a two-week trial and read the log

For two weeks, read every draft before it goes out and every line of the morning summary. You are calibrating two things: the tone of the drafts and the definition of "pending." At the end of the fortnight, adjust the instruction in plain words — "follow-ups should sound warmer," "treat quotes under a week old as warm, not pending" — and the Dot carries the correction forward. This loop is the whole skill.

What to get right before you hand it real work

A few realities from the launch coverage are worth a business owner's attention.

• Permissions are the product. Each Dot runs on its own cloud computer with saved credentials for the sites you connect, and OpenAI says those passwords are stored without being exposed to the model. Treat app connections like keys to the shop: grant the minimum, review them monthly, and disconnect anything the Dot no longer needs.
• Know what your data trains. According to launch breakdowns from ExplainX and The Web Tier, Business, Enterprise and Edu workspaces are not used to improve OpenAI's models by default, and personal plans include user controls for training use. Read the settings for your plan before connecting client data — this takes five minutes and is the cheapest insurance you will buy.
• Keep a human on anything that touches money or reputation. Drafts to customers, invoice figures, refund decisions: the Dot prepares, you approve. The launch coverage is explicit that background research is read-only; keep it that way by policy, not just by default.
• Think about where this is heading. OpenAI is already previewing "specialist dots" — agents with their own identity and credentials built for one job, such as procurement, invoice processing or customer support — and according to TechCrunch's report of the announcement, it is working with Microsoft to manage them under Agent 365 governance controls. The general-purpose Dot you can set up today is the training ground for the job-specific agents arriving next.

The bigger picture: every lab shipped the same idea this September

Dots did not arrive alone. Meta launched Muse — its own always-on agent, each user's running in its own virtual machine — on September 8, per The Web Tier's September launch comparison; and a September LinkedIn recap by Martin Zialcita Jr. reported a Small Business edition landing later that month. xAI's Grok Bot entered public beta in August, each bot getting a persistent cloud computer, according to The Web Tier's September launch comparison and the same LinkedIn recap. The recap's point was that the pattern had converged: strip off the logos and every lab shipped the same idea — an agent with its own computer that runs around the clock, remembers you, and acts while you sleep.

That convergence is the signal worth acting on. The industry is moving from prompting to delegating: instead of getting better at asking, businesses are getting better at describing jobs. A small business that learns to write a one-paragraph job description for a Dot today is learning the management skill every one of these platforms will reward tomorrow. If the idea of handing real customer work to an agent feels premature, the custom AI agents page at https://agenticlabsindia.com/solutions/custom-ai-agents describes how businesses are already scoping job-shaped agents with a human in the loop.