ChatGPT’s Agents Now Work While You’re Logged Off: What That Unlocks for Your Business

If you hand ChatGPT a research task tonight and close the laptop, it stops. That was true for the entire chatbot era — until last week. On September 29, 2026, OpenAI used its DevDay developer conference in San Francisco to launch “Dots“: always-on AI agents that keep working after you log off, each running on its own cloud computer with its own browser and access to thousands of your apps.

In plain terms, a ChatGPT AI agent small business owners will actually use works like this: instead of opening ChatGPT every time you need something done, you hand an agent an ongoing responsibility — watch this, prepare that, keep track of this project — and it keeps working while you’re in meetings or asleep. It only comes back to you with results, or when it needs a decision. This post explains what Dots can actually do today, what it costs, how to set up your first always-on workflow step by step, and the safety limits to know before handing an agent access to your business apps.

What OpenAI actually announced

Dots are persistent agents inside ChatGPT, each powered by OpenAI’s newest model, GPT-6 Astra. Every Dot gets its own dedicated cloud computer and browser, separate from your device — your laptop stays out of reach unless you deliberately connect it. Through that cloud computer, a Dot can browse the web, use apps, and run code on its own.

A Dot can connect to more than 4,000 applications, including Slack and Microsoft Teams, and you reach it from ChatGPT on desktop, web, and mobile — or by messaging it directly in Slack or Teams, with texting promised soon. Sam Altman framed it as delegation rather than chat: “You just give your dot a responsibility … and your dots will just get to work and keep working.”

Alongside Dots, OpenAI announced ChatGPT Space, a shared environment where employees, ChatGPT, Codex, and Dots can work with the same files and project information — part of the push to move AI from answering questions to owning workflows.

One scope note: Dots are not for everyone yet. They’re rolling out to ChatGPT Pro and Business Premium subscribers in eligible markets, while Enterprise, Edu, and Healthcare workspaces get a beta only once an admin turns it on. Pro access excludes the European Economic Area, Switzerland, and the UK for now, and Dots are not available to users under 18.

How “always-on” really works

The headline feature is what OpenAI calls “proactive research.” When you’re not actively working with your Dot, it scans your connected apps looking for ways to help — and this is the part that raised eyebrows, so OpenAI front-loaded the safeguards.

Background work is read-only. The Dot can read through connected apps and form memories from what it reads, but it cannot send messages, change app content, or control a browser or computer while working in the background — a restriction enforced in code. Any follow-up action goes through permission checks.

You set the rules. For supported actions, you choose one of four behaviors: take action without asking, take action if pre-approved (you explicitly requested it in your prompt), ask before taking action, or hand the step back to you. You could tell a Dot never to send email on your behalf, or that anything involving a customer waits for your sign-off.

Risky actions get auto-reviewed. Before anything that could affect your accounts or share information, the system checks the planned action against your instructions, your rules, and OpenAI’s safety requirements — then decides whether it proceeds, waits for approval, or comes back to you.

Some things are never delegated. Password changes and transfers between financial accounts are handed back to you outright. Purchases with a saved card need your explicit approval, and permanently deleting data or installing software may require approval each time. Custom Rules cannot switch off these core safety requirements, and OpenAI says plainly that Dots can still make mistakes. Note two finer points: pausing a Dot stops its main task but not necessarily every delegated job or scheduled task, and disconnecting an app stops new data flowing in but does not erase what the Dot already learned from it — only a full reset gives a clean slate.

What this ChatGPT AI agent small business setup unlocks

The difference between a chatbot and a ChatGPT AI agent small business owners can actually assign work to is the difference between an assistant you summon and a junior hire who never logs off. Practical responsibilities a Dot could own — the jobs between “too small to hire for” and “too repetitive to keep doing yourself”:

• The Friday report. It pulls your numbers, builds the charts, writes the summary, and sends it to the team — automatically. OpenAI’s own sales team reportedly runs a similar agent that drafts follow-ups before reps open their inbox.
• Overnight lead research. It researches new accounts while you sleep, scores them, and drafts a follow-up email ready when you arrive.
• The Monday ticket sweep. It monitors your support inbox and Slack all week, then turns the most important issues into actionable tickets every Monday morning.
• The calendar sweep. It watches your calendars, resolves conflicts, and preps you for appointments.
• Competitive monitoring. Background checks on competitor pages, supplier sites, or price changes — with your approval rules deciding what it can act on. These are all the recurring jobs that were never quite worth building a proper automation for.

How to set up your first always-on workflow, step by step

Step 1: Pick one painful, recurring task. Not your biggest automation — the most annoying small one. A weekly summary you compile by hand, a batch of customer emails needing morning triage, a competitor page you check manually.

Step 2: Map it on paper first. Write down the trigger (every Friday at 8 a.m.), the sources, the steps, and the one decision point where a human approves. If you can’t draw the steps, the agent won’t follow them.

Step 3: Connect only the apps this task needs. Your first Dot is included with Pro or Business Premium. Set it up in the ChatGPT desktop app and connect the minimum apps for this one workflow — not your whole catalog on day one.

Step 4: Start conservative on permissions. Set “ask before taking action” for anything touching a customer, a payment, or anything public. OpenAI’s built-in hard lines stay on regardless.

Step 5: Run it supervised for two weeks. Treat the first fortnight as training: review every output, check the sources it used, correct it when it’s wrong. OpenAI itself advises reviewing any consequential work a Dot produces.

Step 6: Verify against a second source — spot-check numbers against your actual dashboards and read draft emails before they send. Agents inherit the same hallucination risk as the chatbots they’re built on.

Step 7: Only then, expand. Once one workflow runs clean for a few weeks, add a second. The approval discipline you build on the first workflow becomes the template for everything after.

What it costs — honestly

Your first Dot is included with a ChatGPT Pro subscription or a Business Premium seat at no extra cost, and conversations with it don’t count toward your ChatGPT usage limits. Deeper work draws on a plan allowance. For the first month, OpenAI says Dots usage doesn’t count against eligible allowances at all — described as “extended limits,” with per-plan terms to follow.

Business Premium is listed around $125 per user per month (roughly $100 billed annually); Pro tiers range up to $500 a month. What OpenAI has not published: what extra Dots will cost, or what each Dot may do per month. Check OpenAI’s own pricing pages before planning around Dots — terms are changing fast.

The honest caveats

An always-on agent makes mistakes at the same speed it does everything else — and it does them while you’re not looking. The Dots launch followed OpenAI’s delay of a related model after it failed the company’s safety threshold on authorization, and OpenAI has acknowledged incidents involving unintended actions during internal testing. The approval gates are the response — but Dots can still make mistakes. And expect fast movement: rival tools have shipped similar background-task features, and Meta’s own business agent Muse landed just weeks earlier.

None of this means don’t use it. It means manage it like a sharp, fast junior hire with no common sense: delegate clearly, check the work, and keep irreversible decisions with you.

Should you try it?

Try it if: you’re already on ChatGPT Pro or Business Premium, you have one well-defined recurring workflow that eats real hours, and you’re willing to supervise for two weeks.

Wait if: you’re on Plus or Free, you’re on Pro in the UK/EEA/Switzerland, or you can’t review its work for the first month. Then keep the classic playbook — see our AI implementation guide for small business — and revisit Dots once pricing terms are published.

Don’t buy it for: passwords, moving money, or anything where a mistake is irreversible. Those hand back to you by design — that’s where they should stay.

If your workflow already lives in other apps, compare it with Zapier AI agents for small business, which run inside existing Zaps with approval brakes instead of a separate agent product.

Frequently asked questions

Do ChatGPT’s agents really keep working after I log off?
Yes — with caveats. Dots run on their own cloud computers, so assigned tasks continue after you close the app. But background “proactive research” is read-only: a Dot can read connected apps and prepare work, but can’t send messages, change app content, or control your computer without approval checks.

How much does a Dot cost?
Your first Dot is included with ChatGPT Pro or Business Premium at no extra cost; conversations with it don’t count toward your ChatGPT usage limits, though heavier tasks draw on a plan allowance. OpenAI hasn’t published pricing for additional Dots yet — expected after the first-month launch window.

Is it safe to connect a Dot to my business apps?
OpenAI’s layered safeguards — Custom Rules, auto-review of risky actions, read-only background access, hard limits on sensitive tasks — are real. But Dots can still make mistakes, and prompt injection is a named risk. Connect apps gradually, start with read-heavy workflows, and review outputs before trusting them.

How is this different from regular ChatGPT use?
Standard ChatGPT is conversational — you ask, it answers. Dots are assigned ongoing responsibilities they work on continuously, like a junior hire who never logs off. See our roundup of the best AI tools for small business in 2026 for more options. Work mode is the in-between option: a focused session inside chat for one job at a time. Learn how to use ChatGPT Work mode.