By John Wheeler, founder of Wheeler Intelligent Systems LLC
Most of us have learned to use AI through a simple exchange. We ask a question, get an answer, and decide what to do with it. We might ask for a draft, upload a spreadsheet, or get help thinking through a problem. When the conversation ends, we usually carry the work forward ourselves.
OpenAI’s Dots introduce a different idea: an AI agent that can hold onto a goal and continue making progress between conversations.
OpenAI describes a dot as an always-on agent in ChatGPT, with its own cloud computer, access to the apps you choose to connect, and memory that supports ongoing work. It can bring results back for review and turn to you for decisions. [1]
For a business owner, that raises an interesting question. What changes when AI can help keep work moving after you step away?
From Asking Questions to Assigning Responsibility
There is a difference between asking for an answer and assigning a responsibility.
“Give me some ideas for improving customer service” is a request for advice. “Review these approved customer questions each week and prepare a report on the most common concerns” describes an ongoing job.
The second request has a purpose, a source of information, and an expected result. It also creates a reason to return to the work over time.
My view is that this shift could be more important than the appearance of the agent or the way we chat with it. Businesses have plenty of questions. They also have work that needs steady attention.
A useful agent would help close the gap between knowing something should happen and getting it done.
Why This Matters to Small Businesses
In a large company, a new project may have a department behind it. Someone studies the issue, someone prepares the material, and someone follows up.
In a small business, that work often lands on the owner.
The owner may know the website needs clearer answers. They may want to study a new market, compare suppliers, or organize scattered notes. Those jobs compete with serving customers and managing daily problems.
An agent that helps prepare useful work could expand the owner’s capacity. The opportunity is especially appealing for tasks that matter but rarely feel urgent enough to make the top of the day’s list.
That does not mean a dot can run any business on its own. It means there may be a new way to get more of the right work ready for human judgment.
What OpenAI Says Dots Can Do
OpenAI’s documentation describes background research, scheduled reminders, recurring checks, and work with permitted connected information. You can review activity through the dot’s profile, including work in progress, scheduled tasks, and completed tasks. Access to your own computer is optional and starts turned off. [1]
These are product capabilities. They do not tell us how well a particular business task will turn out.
A recurring task might produce a helpful report, or it might repeat information that no one needs. A research task might uncover something valuable, or it might return a long summary with little practical use.
The business still needs to define the job and judge the result.
“Always On” Needs a Careful Reading
The phrase “always on” can make people picture an agent constantly taking action across all their systems.
OpenAI makes an important distinction. Its proactive research tools can read permitted information and save private notes, but they cannot directly send messages, change content through plugins, or control a browser or computer. Follow-up actions must pass the usual action rules and safety checks. [2]
That distinction helps set realistic expectations. Finding something worth doing and having permission to do it are separate matters.
For business owners, the practical question is whether the agent has the information, access, and authority needed for the assigned work. If one of those is missing, the work may stop or return for a decision.
Start With a Job You Can Check
For a first experiment, I would choose work with a clear result and a manageable cost if it goes wrong.
Imagine a small retailer considering a new product category. A possible assignment would be to research public information about that category, prepare a short comparison, and identify questions the owner should ask suppliers.
That is a hypothetical use case, not a claim that every dot will complete it well.
The owner can inspect the sources, check the reasoning, and decide whether the report helps. No purchase needs to happen for the experiment to provide value.
Starting this way gives the business a chance to learn what the agent does well before relying on it for actions that affect customers or money.
Better Instructions Begin With Better Business Questions
An unclear assignment makes useful work harder to recognize.
“Help my company grow” leaves almost everything open. The agent could produce marketing ideas, research competitors, draft content, or suggest new products. Any of those might sound reasonable without serving the owner’s current need.
A more useful question might be: “Which parts of our public service description are unclear to a first-time buyer?”
That gives the work a narrower purpose. The result can be checked against the company’s actual services and the questions customers ask.
Owners do not need to become experts in writing prompts. They do need to explain the problem, provide sound information, and describe what a useful answer would help them decide.
Give the Agent Accurate Information
Business knowledge often lives in people’s heads.
The owner knows which jobs are a good fit. A salesperson knows why customers hesitate. An experienced employee knows which promises are safe to make.
An agent cannot reliably use facts it has never received. When information is missing, a polished answer may hide the gap.
Before assigning work, gather the relevant material. For a website review, that might include approved service descriptions, operating areas, and common customer questions. For a report, it might include definitions of the fields and the period being reviewed.
This preparation has value beyond AI. It makes the business easier for people to understand too.
Set Boundaries Around Actions
A research mistake and an action mistake can have very different effects.
A weak draft can be revised. A message sent to the wrong customer may be harder to fix. A purchase or change to a live system may create costs that extend beyond the original task.
OpenAI describes approval requirements, Custom Rules, and automatic review of certain planned actions. It also warns that dots can still make mistakes and that advance approval remains limited to what the user authorized. [2]
My recommendation is to match freedom to the task. Give an agent room to prepare low-risk work, while keeping clear approval points around important business actions.
The goal is to make useful progress possible with a level of control the owner understands.
Be Deliberate About Connected Data
Connecting an app deserves more thought than switching on a convenient feature.
OpenAI says plugin connections are shared across dots, ChatGPT, ChatGPT Work, and Codex. It also says disconnecting a service stops new access but does not remove information already retained in a dot’s context. [2]
For a business, I would begin with the information needed for the first task. A public research assignment does not need the company’s entire customer history.
This is also a good reason to separate examples from real records. Fictional data can help test a process without bringing customer or employee information into the experiment.
Access should grow because a useful task requires it, with the owner aware of what is being connected.
Measure Useful Results
An agent can produce a lot of activity. That activity needs to lead somewhere.
For an early trial, I would ask whether the result was used, how much checking it required, and whether it helped a decision or finished a piece of work.
A report that takes ten minutes to review and saves an hour of preparation may be valuable. A report that requires two hours of corrections may point to a poor assignment, weak source material, or a task that needs more human expertise.
Count the whole process. Include the time spent preparing information, giving direction, reviewing results, and making repairs.
That provides a more honest picture than measuring how quickly the first draft appeared.
Availability and Cost Will Keep Changing
As of October 2, 2026, OpenAI describes a gradual rollout to eligible Pro and Business Premium users, with an administrator-enabled beta for Enterprise. Its current documentation says the first dot is included in Pro or Business Premium, with an allowance for deeper work and extended limits during the first month after launch. Check the current terms before planning ongoing use. [1]
For a business trial, the broader cost question includes human attention as well as the subscription.
The useful measure is what the business gains after review time, corrections, and usage costs are included.
What I Find Most Interesting
What interests me about Dots is the chance to explore a more lasting working relationship with AI.
A single conversation can be useful. Ongoing work adds new questions. Can the agent keep the goal clear? Can it recognize when information is missing? Can it report a blocker plainly? Can it return something a person can actually use?
Those questions deserve practical tests.
I see Dots as an invitation to learn where ongoing AI assistance fits in a business. Some uses may prove valuable quickly. Others may need better instructions, better information, or more capable tools.
The strongest results will come from owners who are willing to test the idea and judge what happens.
Begin With One Useful Responsibility
You do not need to redesign your company to explore this.
Choose one job that has been waiting for attention. Make the expected result clear. Provide the information it needs. Decide which actions require approval. Then review the outcome and improve the assignment.
That could be enough to learn whether a dot helps your business move forward.
OpenAI Dots make the idea of ongoing AI assistance more concrete. Their lasting value will depend on whether that assistance produces work worth using.
Sources
[1] OpenAI: Getting started with your dot
[2] OpenAI: Dots privacy, security, and safety FAQs