Is ChatGPT “Dots” Worth Upgrading to Pro for?

Is it worth upgrading? Ask yourself these questions:

If you're already doing work in ChatGPT, you can do your autonomous agent work in the same app. That's really nice.

Personal Pro plans start at $100 USD per month. Conversations with your dot do not count toward regular ChatGPT usage limits, but tasks it starts or manages in Work or Codex count toward those products' limits. OpenAI documents an allowance for deeper work and extended launch limits; it does not publish an exact dots token budget on the access page. Check your account's usage as you test larger jobs.

Will you use it enough to justify the subscription? Test it on a real brief and judge the quality of the PDF and how much correction it needs.

Open dots in ChatGPT and follow the introduction. You can connect apps during setup or add them later. When the desktop app offers access to your computer, leave that connection off for this first test. Give your dot a name and customize its appearance if you want. See the official setup instructions.

Your dot has its own cloud computer for research and document work. We want it to work while we work on our computer separately. You can add local access later when a task needs your files or apps. A connected local computer needs to stay online with ChatGPT open. Cloud and local computer access are separate.

Review your existing connections, too. Our dot suggested making a thumbnail template based on a Slack message it found through an existing connector. It does piggyback off your existing connectors.

If a reference file does not come through a connector, upload it directly. We said, “Here's the full image because it had trouble pulling it from Slack.”

Pro tip: Skipping new connections does not remove existing app access. Review each connected account and its allowed actions. Use read-only permissions where available if you only want your dot to research.

Set your operating boundaries before assigning the job.

Set some rules:

I want to set some operating boundaries. I don't want you to ever send a message on Slack from my account or an email from my account without asking me first.

I don't want you to delete any documents created by humans in Google Drive without asking me first, nor do I want you to delete meetings created by humans or invite people to meetings created by humans without asking me first.

Also, I want you to keep your computer organized so that you can use it as a workspace that scales with the projects we work on.

It confirmed a custom rule during the conversation.

For ongoing controls, open Settings > Personalization > Permissions > Custom rules and review what is saved. App permissions and built-in approval checks also apply; custom rules are optional. See dot controls.

Working in PDFs is really hard for agents sometimes, so we always like to ask them to do that because people work in PDFs. They're great.

We asked our dot to do some research and write a proposal with three options for automating thumbnail ideation.

Try giving your dot a task like this:

Can you do some research and write a proposal with three options for automating thumbnail ideation?

We can use ChatGPT image. We can use any image model through OpenRouter to generate a background. I have cutouts that we can put over top. I need you to decide on how we're going to do things like text and templating and layouts. I want it as automated as possible with scripts.

The only open-ended thing will be taking a video title or description and then turning it into a prompt for the background and a layout choice. It's all going to be human reviewed.

This is for ideation and a good first draft. You can do research on how, or if this is already solved, what libraries might be good. I want a PDF with this report in it with links and citations and maybe images even.

Let it work. It is kind of hard to tell what it's working on sometimes, but you can interrupt it and just ask for a status update like a real person.

Open the files your dot returns and inspect the actual document. It took roughly 10 minutes from the research request to opening the PDF. These citations are really thorough. These layouts, they're not bad. These are just the proposals.

Check the facts:

• Does it compare three workable options for the task you gave it?
• Do the links support its recommendations, and do current prices or capabilities match the original sources?
• Do the diagrams explain the workflow clearly?
• Is it clear what can be automated and what still needs your review?

You can inspect delegated work through your dot's profile under Activity. We also opened the cloud file tree using the + button. For browser work, Take over lets you use the cloud computer yourself, then return control when you are done.

Pro tip: Pause stops your dot's current main task. Delegated tasks in Activity and recurring work in Scheduled need separate stops. Pausing does not undo completed actions. See how to stop work.

We were babysitting it, but we didn't have to.

We're really impressed. It could be a no-brainer to upgrade if you're already on a ChatGPT lower plan.