What Are ChatGPT Dots? Always-On AI Agents Explained
Do you still re-explain the whole situation every time you hand work to AI, then tell it what to do next as soon as it answers? Even when the answers come fast, keeping the work moving stays your job.
An always-on AI agent is an AI that remembers the goal you gave it and keeps working on it between conversations, instead of stopping after a single answer. ChatGPT dots, which OpenAI announced on September 29, 2026, are the clearest example so far. The daglo team put this guide together from what we heard at OpenAI DevDay Seoul 2026 and from OpenAI's official documentation.
What are ChatGPT dots and how do they work?
A dot is a personal agent inside ChatGPT that keeps working toward a goal you set, even after the conversation ends. It handles most steps on its own and checks in with you only when it needs to.
According to OpenAI's announcement, dots work like this:
- Their own workspace: Each dot has its own cloud computer and browser and uses the apps you connect. Your own computer stays separate unless you connect it.
- Rules for acting alone: Dots start with default rules for when to act and when to ask first. You can add custom rules that allow an action, require your approval, or block it.
- Sensitive tasks stay with you: Things like changing a password always remain in your hands.
OpenAI's getting started guide notes that dots are rolling out gradually to certain paid plans and regions, so your account may not have access yet.
Chatbots vs. AI agents vs. always-on agents
The difference comes down to who starts the work and when it ends. A chatbot stops after it answers, a typical agent stops when one task is done, and an always-on agent keeps going as long as the goal is open.
| Chatbot | Typical AI agent | Always-on AI agent | |
|---|---|---|---|
| What starts the work | Your question | A task you assign | An ongoing goal and new information |
| How long it works | One reply | Until the task is done | Until the goal is met |
| Memory of progress | Within one chat | During the task | Across conversations |
| Your role | Prompt every step | Kick off and review | Set goals and permissions, approve key decisions |
At the Seoul event, presenters framed this as a shift in how people use AI: from one question and one answer to ongoing projects and real responsibilities. They also suggested that agents will increasingly be judged on autonomy, how often they actually finish the work, and how well they keep track of where things stand, not just on the quality of a single reply.
What always-on agents can do: examples from the keynote
The Seoul keynote walked through cases where a dot takes on work that spans days and moves to the next step whenever something changes. In every case, nobody has to re-prompt it to keep things going.
- Visa applications: spotting details that don't match across documents, then following up.
- Codebase upkeep: finding every place that still uses an outdated API, fixing it, and opening a pull request.
- Status tracking: keeping an eye on bug reports and recurring operations work.
- Pulling services together: combining information from calendars, subscriptions, and other services into a single piece of work.
The keynote also introduced ChatGPT Space, a shared place for documents, files, pages, and agents, and Specialist Dots, which take on dedicated roles inside a company. According to OpenAI's announcement, specialist dots are currently in a focused enterprise pilot.
Three things to decide before you hand work to an AI agent
An always-on agent keeps moving while you're away, so decide up front how far it can go alone. Settle three things first: what it can access, what needs your sign-off, and how you'll check its work.
- Access: Choose which apps and folders to connect. Start with read-only access and widen it only where needed.
- Approvals: Require sign-off for anything hard to undo, such as sending emails, making payments, sharing externally, or confirming meetings.
- Verification: Review the agent's activity log and check numbers and dates against the source.
Work marketers and planners can hand off
You don't need to code to get value from an always-on agent. The best fit is work you check regularly, across several places, where a person only needs to make the final call.
- Collecting weekly changes to competitors' announcements and pricing pages into a short summary
- Gathering campaign results into a first draft of the weekly report
- Comparing an event prep checklist against the schedule and flagging anything that's slipping
- Tracking how far the action items from last week's meeting have moved
If one agent isn't enough and you want to split work across several, read our guide to multi-agent systems.
Agents can only pick up where your meetings left off
For an agent to keep work moving, it needs a record of what was decided and who agreed to do what. Much of that context is spoken in meetings and calls, so a transcript is where the agent's work begins.
daglo transcribes and summarizes your meetings and calls, and keeps the full transcript next to the summary. Ask Board Chat something like "List only the action items we agreed on in this meeting," and it answers from the record, giving you goals and tasks you can hand straight to an agent. If you've noticed AI forgetting the start of a long chat, our post on compaction explains why.
Knowing who said what matters too. daglo's Speaker ID remembers the voices of the people in your meetings, so from the next recording on it fills in their real names instead of "Speaker 1." daglo is the first in Korea to offer this to every user. When the names are right, an agent won't mix up whose task is whose.
#AIagents #AlwaysOnAgents #ChatGPTDots #Dots #ChatGPT #OpenAI #Productivity #daglo
Where talk becomes work.
Get started with 1,000 credits a month — no credit card required.