Trust and quality notes
- Last updated
- September 30, 2026
Dots for Business are OpenAI’s move from an assistant you open when you need help to an agent that can keep responsibility between conversations.
A dot can stay aware of an ongoing goal, use connected apps and its own computer, work in the background, and ask for attention when a decision or sensitive action needs a person. That makes dots potentially useful for the small jobs that are important enough to remember but too scattered to justify a custom automation project.
It also gives the agent more access and more room to make mistakes. The right way to evaluate dots is not by asking how many tasks they can attempt. Ask whether one dot can own one repeated business outcome with clear permissions, visible work, and reliable review points.
What are OpenAI dots?
OpenAI describes dots as always-on agents that can handle ongoing work. Each dot has a cloud computer, can use connected apps, and builds context from the work and feedback it receives.
OpenAI distinguishes a primary dot, which works for an individual user, from a specialist dot, which owns a dedicated organizational responsibility with its own identity, credentials, and system access. Specialist dots are currently offered through focused enterprise pilots rather than normal self-serve setup.
Unlike a normal ChatGPT conversation, a dot is meant to remain present. You can give it a standing responsibility, message it through ChatGPT and supported communication channels, and let it continue authorized work while you are doing something else.
OpenAI says dots can work across text, Slack, and Microsoft Teams. Workspace availability and controls vary by plan, region, and administrator settings, so a feature shown in a launch example may not yet be enabled for every account.
What can dots do for a business?
Dots are broad by design. The useful business applications are the ones where “always on” solves a real coordination problem.
Keep a project moving
A project rarely stalls because nobody can write a status update. It stalls because decisions, messages, files, and next steps are spread across several places.
A dot can monitor the connected context, maintain a working view of the project, prepare follow-ups, and surface the point where a person needs to decide. This is especially useful for work that continues for weeks rather than ending after one prompt.
Keep the responsibility specific. “Help with the launch” is vague. “Maintain the launch action list, prepare the Monday status, and flag blocked work with its source” is a job that can be checked.
Follow up on routine commitments
Invoices, meeting actions, renewals, approvals, and customer follow-ups often slip because they do not belong to one focused work session. A dot can watch the permitted sources and keep a review list of items that need attention.
Start with reminders and drafts. If the dot learns to classify the work reliably, decide later whether any low-risk follow-up can happen with lighter review.
Prepare a recurring brief
A dot can gather permitted information, use connected apps, and prepare a regular brief. Examples include a weekly pipeline summary, an account health review, a competitor update, or a morning list of decisions that need attention.
A useful brief shows where its information came from. Require links, attached evidence, or clear references to connected records. If the dot cannot trace an important claim, it should mark the gap instead of smoothing it over.
Work across chat and company tools
OpenAI’s workspace admin guide describes controls for Slack, Microsoft Teams, local computers, cloud computers, apps, and custom rules. Only a dot’s owner can direct it, although its posts in shared channels can be visible to other people.
That distinction matters. A dot can participate where work happens without becoming a shared command surface for everyone who can mention it.
Delegate parts of a larger task
OpenAI says a dot can perform tasks and delegate work to other agents. The user can inspect current and delegated work in Activity View, provide more context, correct misunderstandings, change direction, or ask the dot to stop.
This is useful when a job contains research, document work, and computer actions. It also makes observability essential. If a business cannot see what is running, what finished, and what evidence shaped the result, it cannot supervise the work responsibly.
Dots pricing and availability
At launch, the first dot is included at no extra cost for eligible ChatGPT Pro and Business Premium users. Enterprise, Edu, and Healthcare workspaces can receive beta access when an administrator enables it. Availability varies by market, and some functions are still rolling out.
OpenAI lists ChatGPT Business Premium at $100 per user per month when billed annually or $125 per user per month when billed monthly. A Business workspace requires at least two Standard or Premium seats. OpenAI has not published a separate standalone price for a dot, and says more dots and higher work capacity are planned.
The price of the underlying plan is not the full cost of a dot. Deeper work can draw on plan allowances, and tasks a dot starts in products such as Codex or ChatGPT Work may use those products’ limits.
For a business, measure the cost of the completed workflow. Include subscription or usage charges, setup time, review time, and correction work. A dot is valuable when it removes a real responsibility, not when it merely creates more output to inspect.
The controls a business should understand
A dot can use sensitive context, apps, computers, and browser sessions. OpenAI’s privacy, security, and safety FAQ describes several layers of control.
App permissions are shared
Plugin permissions are shared across dots, ChatGPT, ChatGPT Work, and Codex. Disconnecting an app stops new access through that connection, but it does not remove information already incorporated into a dot’s context.
Review existing connections before setup. A company may discover that an account already has broader app access than the first dot needs.
Sensitive actions require stronger review
OpenAI says the most sensitive actions, such as changing a password or transferring money, require the user to take over. Other consequential actions, including permanent deletion or software installation, may require approval each time.
Custom Rules can add boundaries, but they cannot override OpenAI’s built-in safeguards or required review checks. Write rules for the business process anyway. “Never send a customer message without approval” is clearer than relying on the agent to infer which action matters.
Activity View makes work inspectable
Users can see tasks, status, and steps in Activity View. They can correct the dot, provide more context, redirect it, or stop the work.
Inspection should be part of the workflow, not something used only after a problem. Review early runs closely enough to understand where the dot needs stronger instructions or less access.
Data treatment depends on the workspace
OpenAI says content from ChatGPT Business, Enterprise, and Edu workspaces is not used to train its models by default. Personal plan settings differ. OpenAI also says content is encrypted while stored and while moving between the user, OpenAI, and service providers.
Those statements are useful inputs, not a complete security review. A company still needs to classify the data involved, check retention and access requirements, and decide whether the connected systems are appropriate for the job.
Important limitations
Dots can make mistakes. OpenAI explicitly says users should review consequential work, and that protections against malicious instructions reduce risk but do not remove it.
Users currently cannot inspect, directly edit, or delete individual dot memories. Deleting a dot removes its context, but separately stored files, Codex threads, or ChatGPT conversations may remain under their own retention rules.
Cloud computers do not automatically inherit a user’s local virtual private network, browser sign-ins, or device policies. Enterprise phone messaging through iMessage, RCS, or WhatsApp was unavailable at launch. Some features may require separate administrator controls or app approvals.
These limits do not make dots unusable. They tell you where to place the first workflow: low enough in consequence to supervise safely, but valuable enough to justify improvement.
A strong first dot for a business
Use a responsibility that is ongoing, evidence-based, and easy to review.
A weekly client delivery coordinator is one example:
- Give the dot access only to the approved project channel, calendar, and task source.
- Define the desired output: completed work, blocked work, decisions needed, and next commitments.
- Require a source reference for every status claim.
- Let it draft follow-ups, but do not allow external sending.
- Review the Activity View during the first runs.
- Record missed context, false alarms, and corrections.
- Tighten the rules before adding more access or more clients.
This workflow tests whether the dot can maintain continuity without putting money, customer trust, or company records at immediate risk.
Dots versus a managed Super Agent
Dots give eligible ChatGPT users an always-on agent inside OpenAI’s ecosystem. They are attractive for teams already comfortable with ChatGPT and willing to configure access, rules, and supervision.
A managed Agentic Workers Super Agent begins with the business outcome and is built around the systems, boundaries, and review loop needed to own it. The distinction is similar to buying capable software versus having the responsibility shaped and run as a working service.
Choose a dot when the job fits OpenAI’s available surfaces, someone can manage the setup, and the team wants a self-serve path. Choose a managed agent when the work crosses systems, requires custom coordination, or needs an accountable operating loop rather than another product rollout.
The bottom line
Dots for Business matter because they are built for continuity. A dot can remember the responsibility, return to it, use connected tools, and bring the person back in when the work reaches a decision.
That does not make broad access a good starting point. Give the first dot one repeated outcome, the least access it needs, and a visible review step. If it produces reliable work under those conditions, expand the responsibility. If it does not, fix the job definition before adding more tools.
Sources
- OpenAI: Introducing dots
- OpenAI Help Center: Getting started with your dot
- OpenAI Help Center: Manage dots in ChatGPT workspaces
- OpenAI Help Center: Dots privacy, security, and safety FAQs
- OpenAI Help Center: What is ChatGPT Business?
- OpenAI: How we build safety, security, and privacy into dots
Learn how a Super Agent is built around a real business outcome