Who’s using Dots? Are they connecting the dots for you?

[VeitB]

The part of Dots that caught my attention is that they’re designed to stay around.

Give a Dot a project, a few things to watch, or something you want to move forward, and it can keep that context instead of starting over every time you open ChatGPT.

Introducing dots, always-on agents built to handle everything.

It can work across several projects, check connected information, surface things you might have missed, and come back when it needs your input.

And importantly, always on does not mean constantly burning through your normal ChatGPT usage.

Your first Dot is included with eligible plans. Normal conversations with it do not count against your usual message allowance. Deeper work has its own allowance, while Codex and ChatGPT Work jobs use their respective limits.

That makes the concept much more interesting to me. The Dot can simply remain available, while heavier work is accounted for when it’s actually needed.

You can also control how independently it acts, whether it can proceed, should ask first, or needs you to take over.

For now, Dots are rolling out to eligible Pro, Business Premium, and Enterprise users, with specialist versions for organizations being explored as well. EU and UK will follow soon.

That’s the product description.

What I’m more interested in is how people are actually using them.

If you’re already using Dots, have they started doing anything you didn’t expect? Will you use them for building things, organizing your life, or something else entirely?

How are Dots connecting the dots for you?

[platypus]

Waiting for them to be available in EU. Also very curious to know how people use them. Do they bring a certain “personality” to AI, or are they just another way of interfacing?

[mitchell_d00]

It is very interesting, I know too little about them. But uninstructed background makes me tenuous. All that said I am excited to see what folks do with it. I’m also interested in seeing how authority and control seat

[_j]

If OpenAI didn’t promote the product by depicting talking to some built-in video walls dedicated to a picture of a blob, filmed in a fake sci-fi control room constructed in a studio, one might be interested.

“If you think your users are idiots, only idiots will use it.”

• Linus Torvalds

A while back, I generated and shared an image of an ad featuring a “ChatGPT Store” where you can go pick up your gpt-6 package - just like the mobile phone or insurance ads that have a non-existent clean shopping center with helpful customer service, for fun. It looks like OpenAI ran with the advertisement and marketing fantasy.

[VeitB]

@_j

one might be interested.

Got it! You’re interested and just need a little push.

What would you do with it once your interest has been piqued?

[VeitB]

I have an experience to share:

I set up my Dot this morning and gave it access to email.

It immediately noticed that my flight was delayed and suggested I look into rebooking my connecting flight. I dismissed it.

When the flight was delayed again, it pinged me a second time. And yes, I needed to rebook.

I never asked it to do any of this.
It’s apparently just trying to help wherever it can.

[platypus]

That’s amazing. So does it automatically schedule regular tasks (e.g. to check things) or subscribe to e.g. webhook in its VM? (Cause as they say it’s 24/7, so like a VPS)

[curt.kennedy]

My dot is scheduled to run during a virtual meeting tomorrow to try replicating what Motion is doing, like recording and transcribing the meeting and creating actions. Not sure how it will go, but it is scheduled.

Here is what it told me:

"You don’t need to prompt me. I’m scheduled to check after tomorrow’s meeting, with up to an hour of scheduling flexibility.

I’ll retrieve Google’s transcript, read it, and send you my own summary of decisions, action items, owners, and deadlines here. If it’s still processing, I’ll keep checking; if capture failed, I’ll tell you.

Your part is to join as host through your browser and handle any recording prompt. Motion will run alongside Google for your comparison"

[platypus]

Put your feet up, relax, and enjoy the ride!

[VeitB]

@platypus

does it automatically schedule regular tasks

In this case it’s a more user-friendly way to set up scheduled automations. This morning, it was simply working unprompted, checked my emails for lack of a better task, and realized this could be interesting for me. And it must have done so regularly.

Ultimately, it will be able to look at all kinds of connectors that we as developers and builders provide and ask the Dot to work with them.

[platypus]

It’s mouthwatering. Eagerly awaiting their EU release.

[proflead]

I’m starting to use it. I’m currently experiencing a problem when I try to call my dot from the browser (Ubuntu OS, Chrome browser); it doesn’t answer. I also tried from my mobile phone (iOS, ChatGPT app) and am experiencing the same problem. Is there any specific setting I should turn on?

[unicornoptimist]

I’m early in my experience with Dots, and there’s something genuinely promising here. I already run a team of agents, so I’m interested in what changes when an assistant stays with the work, notices what matters, and follows through. My dot, Dolly, has given me enough of that to want to keep going.

The rough edges have mostly been in execution and recovery:

• A calendar edit repeatedly returned a generic “Bad Request.” The underlying issue was a Google API restriction on Gmail-generated events. Identifying that restriction earlier would have avoided repeated attempts and an unnecessary browser-login detour.
• Creating an in-person event added a Meet link when the conferencing option was omitted. Explicitly disabling it worked, but the connector couldn’t remove the existing link. We resolved it by recreating the event.
• Recovering from these small failures required more of my attention than the tasks warranted. That’s the part I’d most like to see improve: clearer errors, better diagnosis, and less human effort spent getting routine work unstuck.

The encouraging part is what happened next. We tested ordinary calendar creation, renaming, moving, and deletion. All four worked, with each result checked. Local browser access also worked. Dolly has surfaced a real calendar discrepancy, useful news, and a subscription price change, while learning which people and priorities matter most to me.

For me, the measure is human tokens: how much attention does the assistant return, and how much does it consume? There’s already value here. More dependable execution and smoother recovery would make that value compound.

Promising start. A few lessons learned. I’m looking forward to seeing where this goes.

[_j]

@VeitB

What would you do with it once your interest has been piqued?

I’d demonstrate the actual “Dot” usage experience and capabilities when you don’t have a video wall showing a blob character.

You’ll immediately note no “copy text” button on the output, nor “try again” nor editing of input nor copy input text. A degraded interaction surface vs original ChatGPT.

AI output copied from browser DOM

• Web research
  • search_query: Search the web.
  • open, click, find: Read pages, follow links, and locate text.
  • image_query: Find images.
  • screenshot: Capture PDF pages.
  • weather, finance, sports, time: Retrieve specialized information.
• Image creation
  • imagegen: Generate images or edit supplied images.
• Browser and computer interaction
  • getState: Inspect available interfaces.
  • getBrowser, getTab, createBrowserTab: Select a browser or access tabs.
  • get_screenshot: Capture the desktop.
  • click, move, drag, scroll: Operate visible controls.
  • press_key, type_text: Enter keyboard input.
• Files and command execution
  • exec_command: Run shell commands.
  • write_stdin: Interact with a running command.
  • apply_patch: Edit files.
  • download_file: Download an authorized uploaded file.
  • view_image: Inspect a local image.
• Connected resources
  • list_mcp_resources: List available resources.
  • list_mcp_resource_templates: Discover parameterized resources.
  • read_mcp_resource: Read a resource.
• User communication
  • send_message: Send text or file attachments.
  • read_messages, search_messages: Retrieve conversation context.
  • add_reaction, remove_reaction: Manage message reactions.
  • request_location: Ask for a device-location snapshot.

This summarizes the user-facing tool capabilities described at the start of this context. Availability and required permissions depend on the tool and connected service.

Tools are those of web interaction, after no remote control of your computer is allowed. MCP resources, I did not have interest in polling.

You can add any of hundreds of emoji reactions pointlessly to the AI output, though.