On September 29, 2026, OpenAI introduced Dots, always-on AI agents that keep working after the conversation ends. Most AI tools still wait for a person to start the work, give instructions, and come back when the next step is due. A Dot gets its own cloud computer, a browser, access to the apps you connect, and enough memory of your goals to keep working on them.
OpenAI says Dots run on GPT-6 Astra, connect to more than 4,000 apps through its plugin ecosystem, and can be reached in ChatGPT, Slack, and Microsoft Teams. That moves AI closer to something a business delegates responsibility to, rather than something employees open when they need an answer.
AI has been able to write, research, and automate tasks for years. What OpenAI’s Dots announcement changes is how much responsibility an AI system can hold between prompts.
What Are OpenAI Dots?
OpenAI Dots are persistent AI agents that use their own cloud computer, browser, and connected apps to do ongoing work on a user’s behalf.
A Dot can run several projects at once and keeps the context of each one, so you can hand it new tasks without opening separate threads or directing every step. You can message it or call it by voice, and it messages you back with progress, questions, and decisions that need you. It also learns from feedback, so its work drifts toward the way you would have done it.
The difference from a chatbot is persistence. Instead of asking an AI tool to analyze one batch of support tickets, a business could give an agent standing responsibility for support feedback: watching it, spotting recurring problems, investigating them, and preparing fixes.
OpenAI’s own software example works exactly that way. A Dot watches customer feedback for recurring requests, scopes small improvements and bug fixes, builds and tests them, and hands a developer finished pull requests to review, with videos showing the changes.
That looks less like prompting an assistant and more like assigning work to a team member.
Who Can Use Dots, and What They Cost
As of launch:
- Pro and Business Premium ChatGPT plans get Dots first, in eligible markets.
- Enterprise, Edu, and Healthcare workspaces can try a beta once an admin turns it on.
- Your first Dot is included in Pro and Business Premium at no extra cost, with an allowance for deeper work and extended limits for the first month.
- Conversations with a Dot don’t count against ChatGPT usage limits. Tasks it starts in Codex or ChatGPT Work do.
- Later, OpenAI plans to let you add more Dots and pay to make each one faster or able to take on more work per month. It hasn’t published those prices.
You set a Dot up in the ChatGPT desktop app or a desktop browser. After that it’s available in the mobile app too.
Dots Are Part of the Shift From AI Assistants to AI Agents
We covered this shift in our guide to agentic AI for business. The short version: an assistant answers a request, while an agent decides what needs to happen next, uses tools, checks the result, and keeps going toward a goal.
Dots add persistence to that model. Most agents in production today are still transactional. A workflow starts, the agent runs a series of steps, returns a result, and stops. An always-on agent keeps track of an ongoing project and reacts when something changes.
For planning purposes, that changes the unit of automation. Instead of listing individual tasks to automate, a business can start listing areas of responsibility an agent could help run.
What Can OpenAI Dots Actually Do?
OpenAI’s launch examples cover software development, product launches, research, sales, and content production. None of the individual tasks are new. Connecting them into one continuous process is.
Software Development
A Dot can monitor product feedback, investigate bugs, make small changes, run tests, and prepare pull requests. That work normally spans several systems: the feedback inbox, the product backlog, the codebase, the test suite, and the project tracker.
The hardest automation problems usually sit in the gaps between those systems. A support ticket identifies a problem, but someone still has to understand it, decide whether it belongs on the roadmap, write it up, assign it, fix it, test it, and tell the customer. A persistent agent can take part in far more of that chain than a single-task tool can.
Product Launches and Marketing
OpenAI describes a Dot that learns a launch’s audience, positioning, and creative standards. When the product scope changes, it works out what that means for the story, revises the launch materials, and drafts the affected assets and docs for review.
Generating the copy isn’t the new part. Knowing that a change in one place means updates in five others is.
Research and Analysis
For research, a Dot keeps an analysis current instead of producing it once. As new data arrives, it reruns the analysis, investigates unexpected results, updates the figures, and flags what needs a person’s review.
Plenty of businesses rebuild the same report every week or month. Persistent agents could turn some of those reports into systems that maintain themselves.
Sales
OpenAI’s sales example follows the same pattern. During an enterprise deal’s technical review, a Dot checks customer requirements and account history against product documentation, finds what still needs testing, builds a proof of concept for a key integration, and updates the proposal and test plan as requirements change. The sales lead and engineer stay in the customer conversations and approve any commitments.
Proactive Research: Work Nobody Asked For
The part of the launch with the most practical reach is what OpenAI calls proactive research. When you aren’t working with your Dot, it looks through the apps you’ve connected for things that need attention.
OpenAI limits this mode to read-only tools. During proactive research a Dot can’t send messages, change app content, or control a browser or computer. It can only find things and bring them to you.
Even with that limit, it breaks the usual automation model. Most automation today needs a defined trigger:
Trigger → Workflow → Result
A persistent agent works more like this:
Observe → Interpret → Investigate → Recommend or act
Real work rarely arrives as a clean event. A customer goes quiet, a requirement changes, a metric drifts, a deadline gets close, or a decision happens in a Slack thread. OpenAI’s launch post gives one example from an early tester: his Dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it once he approved it.
No workflow tool would have caught that, because nobody would have thought to build a trigger for a forgotten invoice.
Does OpenAI Dots Replace Custom AI Agent Development?
For some use cases, yes, and that’s good for buyers. No business should pay for custom software when an existing platform already does the job.
If you need an agent to monitor email, summarize documents, research a topic, prepare reports, or coordinate work across common SaaS apps, products like Dots will handle more and more of that out of the box.
The need for custom work doesn’t disappear. The boundary moves. Custom systems still make sense when an agent has to:
- work inside proprietary software or talk to internal APIs
- follow company-specific rules and approval chains
- meet strict security or data-residency requirements
- hit measurable accuracy targets, with evaluations to prove it
- run inside your own product rather than as a separate app
OpenAI’s own roadmap points the same way. Its specialist Dots, agents with their own identity, credentials, and access to a company’s systems of record, start as focused enterprise pilots where OpenAI’s engineers work directly with each organization to define the agent’s responsibilities, tools, and review steps. It is also integrating specialist Dots with Microsoft’s Agent 365 so IT teams can govern them with the tools they already use. Deep integration takes hands-on implementation work, even for the company that built the model.
That integration work is what our AI agent development team does, alongside our AI consulting and development and custom LLM development services.
Software has always moved this way. Capabilities that once needed custom development become standard product features, and custom work moves up the stack to harder, more specific problems.
Reliability Is Still the Hard Part
Giving a model a browser, connected apps, and ongoing responsibility multiplies the ways things can go wrong.
Real workflows are messier than demos. APIs fail, websites change, logins expire, information conflicts, instructions are incomplete, and models misread context. We’ve written about why AI automation works in demos but struggles in production. A persistent agent covers more ground for longer, so the same problems have more places to show up.
OpenAI built a lot of controls into Dots:
- Each Dot works on its own cloud computer. Your computer stays separate unless you connect it.
- Dots can sign in to supported sites with saved passwords without exposing the passwords to the model.
- Custom Rules let you allow, require approval for, or block specific actions.
- Auto-review checks actions that could affect your accounts or share information against your instructions and rules.
- Activity View shows what the Dot is doing, including background work.
- Some tasks, such as changing a password, always stay with you.
OpenAI also says plainly that Dots make mistakes and that consequential work needs human review.
The practical design question for any business isn’t how to remove people from a workflow. It’s deciding where an agent can act alone, where it should ask first, and where a person keeps the final decision.
What Dots Mean for Small and Mid-Sized Businesses
Until now, this kind of automation took a real technology stack: developers, APIs, workflow tools, an orchestration layer, prompts, hosting, monitoring, and someone to maintain it all. Dots package much of that into a product that a smaller business can configure without building from scratch.
That changes the first question. Instead of “How do we build an AI agent?”, a business can ask “Which parts of our business should an agent be responsible for?”
Answering that means understanding the workflow before picking the technology. Where does information enter the company? Who reviews it? Which decisions repeat? Where are people copying data between systems by hand? What gets forgotten or delayed? What gets rebuilt every week? Which steps need judgment, and which follow predictable rules?
Those questions surface better opportunities than starting with a product and hunting for a place to use it. It’s also why our AI consulting for small businesses starts with the workflow, not with a model or a platform. Sometimes the answer is a custom system. More often now, it’s configuring a product like Dots correctly and connecting it to the rest of the business.
Where This Is Going
Today you start with one primary Dot. OpenAI says it envisions teams of Dots working together, and its specialist Dots are built to take on defined roles inside an organization. The internal testing it cites covers procurement, invoice processing, email marketing, customer support, and commercial contracting.
Put together, that looks less like a chatbot and more like a software layer running across a company. A marketing agent maintains campaigns and watches performance. A product agent tracks customer feedback and keeps requirements current. A development agent investigates issues and ships small fixes. An operations agent tracks open work, and a reporting agent keeps the metrics up to date.
People still set direction, approve important decisions, handle the unusual cases, and stay accountable for the results. We looked at that balance in whether AI agents replace or augment employees. What changes is how much of the coordination in between an agent can carry.
We aren’t there yet, and the limits will become clearer as businesses put Dots into real workflows. But the direction is now easy to see.
The Bigger Story Isn’t Just Dots
Every big AI launch sets off the same argument about whether it’s revolutionary, incremental, or overhyped. With Dots, the more useful question is what it says about where AI software is going.
AI is moving from something people use toward something businesses delegate work to. Dots are one version of that, and others will follow. Once always-on agents are a standard feature, having one won’t be an advantage by itself.
The advantage will come from knowing your own workflows well enough to decide what to delegate, what to integrate, what can run on its own, and what stays under human control. That’s the conversation worth having now.
