ChatGPT Dots Explained: OpenAI's Always-On AI Agents, and the Jobs They Change First

On September 29, 2026, at its DevDay conference in San Francisco, OpenAI launched dots - always-on AI agents that live inside ChatGPT.

Here is the simplest way to understand them. A chatbot waits for you. A dot does not. You give it a goal, and it keeps working after you close the app: checking, chasing, booking, fixing. When it needs a decision, it messages you.

That small difference - from a tool you use to a helper that works in the background - is why dots may change more jobs than another smarter model would. This guide explains how dots work, who can get them, what they cost, how safe they are, and which jobs they touch first.

What is a ChatGPT dot?

A dot is an AI agent with four things a normal chat does not have:

• A goal it keeps. You set it once, and the dot works toward it between your conversations.
• Its own computer. Each dot runs on its own cloud computer, so it can browse, use tools and do real work - not just write text.
• Its own memory. It remembers the goal, the apps it is connected to and what it has done so far.
• Its own schedule. It sets its own recurring checks and messages you when it needs your judgment.

Every dot runs on GPT-6 Astra, OpenAI's newest model, and can connect to more than 4,000 apps. You can talk to it in ChatGPT, from your phone, or in Slack and Microsoft Teams.

Chatbot vs dot: what actually changes

What can a dot do?

OpenAI's own examples are everyday and practical: scheduling, travel, budgeting, debugging and coding. Picture a dot that watches for a cheaper flight and asks you before it rebooks, one that tracks spending against a budget every evening, or one that keeps an eye on a failing test and fixes it overnight.

For companies, OpenAI described specialist dots - agents given one clear responsibility, each with its own identity, login details and system access, like a new team member with a narrow job description.

Who can get dots, and what do they cost?

The engine: why GPT-6 Astra makes dots useful

An always-on agent is only worth having if it can actually finish the work. That is where GPT-6 Astra comes in.

OSWorld puts a model in front of a real desktop and asks it to complete real tasks - fill in a form, update a spreadsheet, move files between apps. Astra finishes 72.6% of them unattended.

It also does them in about 40 minutes on average, against 75 for the previous model. Now combine the two: a model that finishes seven in ten desktop tasks, running all night, every night, without being asked. That is the real product OpenAI launched.

How safe is it to let a dot act for you?

Giving software your logins is a big step, and OpenAI built in several layers of control:

• Your own computer is off-limits by default. Access to your local machine is optional and starts switched off.
• Background research is read-only. In that mode a dot cannot send messages, change content or control a browser.
• A human must take over for passwords and money transfers. The dot cannot do those alone.
• Actions are reviewed before they run, and you can set custom rules for what a dot may do.

The limits are not perfect, and OpenAI says so: its safeguards reduce risk but do not remove it. In testing reported from the GPT-6 Astra system card, the model respected changed permissions in most - but not all - cases.

One more catch: you cannot view or edit a dot's individual memories. The only full reset is to delete the dot, along with its conversations and scheduled tasks.

Why dots matter more than another model launch

Most AI launches make a tool smarter. Dots change who starts the work.

Until now, AI waited for a person to ask. The person did the "glue work" around every task: noticing something needs doing, chasing the reply, checking the result, booking the next step. Dots are designed to do exactly that glue - continuously, in the background, across thousands of apps.

That is why the jobs most exposed are not the most technical ones. They are the jobs where a large part of the day is following up, checking and coordinating.

Which jobs feel dots first?

The pattern is simple. If your work is "keep things moving", a dot is built to do it. If your work is "decide where things should go", a dot needs you.

The new skill: managing agents

The people who gain from dots will be the ones who are good at delegating to them:

• Writing a clear goal. "Keep my calendar free on Friday afternoons and book client calls into Tuesday" beats "help with my calendar".
• Setting limits. Deciding what a dot may do alone, and where it must ask.
• Checking the work. Spot-checking what the dot did, and catching the four-in-49 moments.
• Owning the outcome. When a dot books the wrong thing, a person still answers for it.

These are management skills, and they are now worth learning even if you manage no one.

What to do this week

1. List your follow-ups. Write down everything you check, chase or book in a normal week. That list is a dot's job description.
2. Move toward the decisions. Dots message a person when they need judgment. Aim to be that person, not the follow-up the dot replaced.
3. Practise delegating. Try an agent on one repeating errand, write the goal carefully, and check what it did.
4. Know your exposure. See which of your tasks are most at risk, and build a plan around the strengths that protect you.

Dots will not be the last always-on agent; they are the first one most people will meet. The question for your career is not whether agents arrive at work - it is whether you are the one directing them.