Most AI tools wait for you. You open the app, type a prompt, get an answer, close the tab. ChatGPT Dots flips that dynamic: the agent keeps running after you walk away, with its own persistent browser, its own scheduled tasks, and the ability to ping you when something worth knowing happens.
This is a genuinely different category of software, and it takes a few minutes of hands-on time before the implications sink in.
What a Dot Actually Is
A Dot is a persistent AI agent you configure once and leave running. It has:
- A cloud-based browser it controls autonomously
- Scheduled tasks it writes and manages itself
- Memory and context pulled from your ChatGPT history and connected accounts
- Integrations (calendar, Slack, email, and others) it can act on proactively
The closest analogy is hiring a remote assistant who works the night shift, checks things on your behalf, and leaves a tidy report on your desk each morning. Except the assistant never sleeps and charges no overtime.
Setting One Up: The Core Pattern
Configuration follows a three-step logic that applies regardless of what job you give your Dot.
1. Give it real context about you
A Dot is only as useful as what it knows. Before you assign any task, tell it how you work: your time zone, your communication preferences, the tools you rely on, any recurring priorities. Think of this as onboarding a new hire, not filling out a form.
If you want it to manage your calendar, for example, explain your meeting philosophy. Do you block mornings for deep work? Do you refuse back-to-back calls? The more specific you are upfront, the less you’ll need to correct it later.
2. Connect the right integrations
Dots become powerful when they can read and write to the places your life actually lives. Link your calendar, your email, your project tracker. A Dot that can only browse the web is useful. A Dot that can see an overdue invoice in your inbox and add a follow-up block to your calendar is genuinely valuable.
Start with one or two integrations, not all of them at once.
3. Define at least one scheduled task
This is where most people underuse the product. A Dot without a schedule is just a fancier chatbot. The real leverage comes from recurring checks:
- Scan a job board every morning and summarize new listings that match your criteria
- Check a supplier’s website for price changes and log them to a shared document
- Review your calendar each Sunday evening and flag any conflicts or missing prep time
- Monitor a shipping status page and notify you the moment something updates
The Dot handles the repetitive surveillance. You handle the decisions.
The Built-in Document Space
OpenAI also shipped a Notion-style document layer inside ChatGPT called Spaces. It’s a Markdown editor where you can create nested pages, embed AI-generated content, and — critically — have your Dot write directly into it.
That last part matters. If your Dot is monitoring something, you want that information somewhere structured, not buried in a chat thread. Telling your Dot to maintain a live tracker inside a Space means you get a document that grows and updates on its own. You open it when you need it; it’s already current.
The collaboration features (comments, sharing) are straightforward and work as you’d expect if you’ve used Notion or Google Docs.
What to Watch Out For
Sites are blocking agents. Amazon, and likely others, have started detecting and blocking automated browsing. This will be an ongoing arms race. For now, the most reliable use cases involve sites that don’t have strong bot-detection — niche marketplaces, government portals, smaller SaaS tools.
Tasks take longer than you expect. Creating a structured document from a web research task can take several minutes. This isn’t a tool for instant results. It’s a tool for results you didn’t have to babysit.
The interface is still maturing. Everything lives inside the main ChatGPT UI for now, which can feel cluttered when you’re managing multiple Dots with multiple schedules. A dedicated management view would help; presumably that’s coming.
The Right Way to Start
Don’t try to automate your entire life in the first session. That path leads to a pile of half-configured tasks and nothing actually working.
Instead: pick one specific pain point. Something you do manually right now that’s repetitive, time-sensitive, and easy to define. Price tracking, meeting prep, status monitoring, inbox triage — any of these are good starting points.
Configure one Dot for that one job. Give it context, connect one integration, set one recurring task. Let it run for a week. Once you see the output arriving without you lifting a finger, the rest of the possibilities will make much more practical sense.
The agents that will save you the most time aren’t the ones doing the most things. They’re the ones doing one specific thing reliably, every day, while you’re focused elsewhere.