OpenAI dots: what the launch means for business AI

What OpenAI confirmed about dots

OpenAI describes dots as always-on agents in ChatGPT. A dot can own ongoing work and keep making progress between conversations. Dots run on GPT-6 Astra and have their own cloud computers. They can use the apps a person chooses to connect, and they bring work back for review when a decision needs human judgment.

It’s more than a longer chat. OpenAI says a dot can carry context over from ChatGPT memory and use Codex and connected tools. It can keep up scheduled work and share updates without being asked. People can see work in progress, scheduled and done. They can change direction, pause the agent or use its computer.

Control is part of the official design. Custom Rules can allow an action, require approval first, ask before acting or hand the step back to the user. OpenAI also says actions that could have serious effects go through automatic review. It warns that dots can still make mistakes and that important work should be checked.

• Available through ChatGPT on web, mobile and desktop. Creating a dot currently starts on desktop web or the desktop app.
• Rolling out gradually to Pro users outside the EEA, Switzerland and the UK. Business Premium has it in supported ChatGPT regions. The Enterprise beta needs an admin to turn it on.
• Slack is supported as a messaging channel, and the launch page also shows Microsoft Teams. Texting has separate limited-beta conditions.
• Access to your own computer is optional and starts turned off. Otherwise, dots use their own cloud computer.

Sources

• OpenAI: Introducing dots
• OpenAI Help Center: Getting started with your dot

Our view: AI is starting to own a job, not just answer

The confirmed facts stop at what OpenAI has announced. Our view is that products like dots may change what people expect from AI at work. Instead of answering one request, an agent owns a defined job, keeps context and comes back with progress you can review.

That doesn’t mean every process should run on its own. Reliability, cost, access, security and the quality of the work around it decide whether it works in practice. The useful business question is narrower: which job can your team define clearly enough to hand over, supervise and check?

A broad goal like ‘automate customer experience’ is hard to test. A small, clear job is a better start. For example: help new customers prepare for one onboarding step.

Write the job before you choose the tools

Picture a customer asking what to prepare before an onboarding call. A useful pilot names the approved onboarding guide and the details the AI should ask for. It also names the action it may take and the issues that belong with a customer success manager.

Then test typical questions. Include simple requests, missing details, conflicting information and questions outside the job. Write down what a good answer or handover should contain before you judge the technology.

This turns an impressive demo into a test you can repeat. It also makes comparing platforms fairer, because every option is asked to do the same job.

Give your business knowledge an owner

An AI that works over time needs information someone keeps current. Someone should own the product details, instructions, policies and examples it relies on, including when each was last checked.

Think about a price change. If one document has the new terms and another still describes the old offer, adding both doesn’t fix the clash. The business needs a rule for which source wins, and a person who keeps it current.

Write down the sources the pilot uses and who looks after them. When something goes wrong, your team can then tell a source problem from an instruction, permission or action problem.

Plan approvals and handovers together

OpenAI’s launch puts controls and approval choices close to the agent. Businesses should apply the same care to the job itself. Decide which actions can happen automatically, which need approval and which must stay with a person.

Then decide which details go with a handover. For an unusual support issue, that might be the customer’s goal, the sources checked, the steps already tried and the open decision. A good handover saves repeated work; it doesn’t just move the conversation.

Review hard cases early. They show where instructions are vague, sources clash or an action limit needs to change.

• Low-risk action you can undo: decide whether the AI may go ahead, and how it records the result.
• Action outside your business, or with serious effects: set the approval point and what the reviewer sees.
• Unclear or off-topic request: prepare a short handover with the details, for the right person.
• Sensitive access: give only the permissions the job really needs.

Measure the work that matters

Choose measures that match the job. For onboarding, check whether the guidance matches your approved documents and whether the customer reaches the milestone. For support, look at open questions, repeat contacts and the quality of handovers.

Also track the human effort needed to correct and maintain it. A polished AI can still be costly to supervise. Your pilot should show the useful work is worth that effort before the job grows.

Start with a small set of measures: quality of finished work, accuracy, exceptions, how often approval is needed and how useful handovers are. Read the conversations behind the numbers, so your team understands why a result changed.

Where Aivah fits

Aivah gives businesses AI employees for customer-facing work in sales, support, onboarding and product training. You define the job and add your approved documents. You choose the model, voice and character, connect the apps you allow and review the conversations.

The dots launch and Aivah reflect related changes in how people hand over work, but they are separate products. Aivah has no announced OpenAI dots integration or partnership. Our advice: start with one customer journey, define the job and the proof, then test with real questions before you expand.

The launch is a good moment to make those decisions. The first useful question is simple: which job can your team define clearly enough to hand over and check?

• Explore Aivah AI employees
• See business use cases
• Plan a business demo

OpenAI dots and business AI FAQ

These answers keep OpenAI’s published product information apart from Aivah’s independent view, as checked on 29 September 2026.

Sources

• OpenAI dots product page
• OpenAI dots setup and eligibility