Halfway through episode 32 of One Shot Show, I opened the chat with my dot to show Dheeraj Sharma something I hadn’t typed: a row of messages that Dots started on its own.
One flagged that my flight had been rescheduled. Another reminded me that a guest writer was waiting on my reply. A third pointed at a sponsor placement I still needed to answer. I hadn’t asked Dots to check any of it. It went through what it could see in my email and calendar, decided those things needed me, and brought them up by itself.
That’s the part of OpenAI’s Dots I keep coming back to. Claude Code, the way I run it, still waits for me to open my laptop and type first. Dots can message me first, and of the always-on agents I use, its nudges are the ones most worth reading.
Dheeraj and I spent the episode going through it. I walked through my own setup, and Dheeraj, who mostly lives inside Anthropic’s tools, asked the questions a Claude user would ask before paying for another subscription to use ChatGPT. This post starts with a quick look at what Dots is, then the proactive nudges, because that’s where Dots shines the most, then the rest of the features and the use cases I’ve been running with it.
One Button to Launch, No System Prompt
OpenAI launched dots at DevDay on September 29, 2026. A dot is an always-on agent inside ChatGPT that runs on GPT-6 Astra and works from its own cloud computer and browser. As of mid-October 2026, it’s available on ChatGPT Pro and Business Premium, where your first dot is included at no extra cost. Plus doesn’t get one. The rollout is gradual and varies by region, so check OpenAI’s help page before you upgrade for it.
Mine is called Slash.
When I open its computer, there’s Chrome, Blender, FreeCAD, 3D Slicer, and a few other apps installed, so Slash can browse and work in those apps without my laptop being awake.
It also picks up every app already connected to my ChatGPT account. Email, calendar, Slack, Notion, and meeting notes from tools like Wispr Flow or Granola all come along. And if ChatGPT already has memory about you, your dot can use that too.
To use Dots requires no setup at all. You press a button to launch your dot, and it just appears. I didn’t write a system prompt or set up instructions first. If you’re wondering where the context comes from, it’s drawn from three places: your conversations with it, the files you add, and the apps you connect. Over time, it writes its own memory of you based on what you ask it to do.
So I talk to it like I’d talk to a coworker, in plain requests.
Where the Nudges Come From
Why do I keep using Dots lately?
The key feature I like is its proactive nudges. Dots can nudge you out of the blue, without requiring you to prompt anything or set a specific time, with the goal of reminding you what needs your attention and offering help to take it off your plate.
Here's one I showed: one morning, it sent me three overnight updates—a sponsorship I needed to respond to, a reminder about the live session with Dheeraj that night, and a flight I had to take that day. But the sponsorship ping stood out to me the most.
OpenAI calls the background part “proactive research.” According to OpenAI’s docs, your dot can read the information it has permission to see, keep private notes about useful findings, and connect new information to your work. It is part of research feature, so it can’t send messages, change your connected apps, or control a browser.
It just reports back to your dot, and your dot brings you a suggestion or a question. Any follow-up action still goes through the permissions and approvals you've set, so the final call stays with us humans.
That’s why the nudges stop at a suggestion or an offer to help. Dots noticed the sponsor email and reminded me. Replying was still my decision.
That’s also why I see Dots as a chief of staff for work, and I think that positioning makes sense. When people use ChatGPT, they mostly use it for their work, so their work files are already there. It makes sense to have an agent on top of all that, one that runs 24/7, can access those files, and taps you on the shoulder when something needs you.
If You Already Use Grok Bot, Here’s What’s Different
Grok Bot runs 24/7 too. I previously wrote about my seven-bot Grok Bot setup, so if you’re already running it, you might be wondering what Dots does differently.
The biggest difference for me is the proactive research. Grok Bot shipped its own proactive feature about 1-2 weeks ago, but in my use so far, it’s nowhere near as good as Dots at picking up what actually needs my attention.
The second difference is the model. Dots runs on GPT-6 Astra, which I find much smarter than Grok 4.7, the model behind Grok Bot. Harder or longer-running tasks that get stuck in Grok Bot often go through on Dots. This is one of the reasons I keep coming back to Dots.
But Grok Bot still has one advantage that keeps me using it, which I’ll get to in the limits section below.
Routines Are the Nudges You Schedule Yourself
Proactive research decides on its own when to ping you. Scheduled tasks let you decide when.
The routine I use most runs every day at 8 a.m. Dots reads all my Codex conversations from the day before and tells me where I am: which tasks need my attention, which threads I should follow up on, and which ones I should look at again. I read the update, then pick up where I left off.
I showed this live. I asked Dots, “Can you read my conversation with Codex for today?” and it returned a clickable button. When I clicked it, you could see Dots going through each thread, with the thread ID, the thread name, and the files generated inside each one. That’s how it worked in my account.
Two other routines I mentioned during the session:
Daily priorities. A routine at 7 or 8 a.m. (or the evening before) that reviews your email, calendar, and task notes, tells you what needs attention, and drafts the tasks for the day. You choose where the output lands: the ChatGPT app, Slack, or email.
AI news every morning. Dots reads my AI newsletters, or it can search the web for AI news from the last 24 hours.
Errands You Can Hand Off by Call or Chat
You can also call your dot. It talks back, and while you’re on the call it can check your email, look things up, or adjust a slide deck.
The best example happened after I booked an Airbnb in another city. To be clear, Dots has no access to my Airbnb account, but it reads my email, so it saw the booking confirmation. I called it and asked it to find the closest coffee shop to the Airbnb where I could work. The call took around nine minutes, and it came back with photos of the coffee shops and a suggestion for which one to go to. I later asked it for the nearest gym too.
Another one, this time in chat: I’m currently staying in Bali and wanted a padel coach, ideally someone who could show me how international players approach the game. I just asked it to find me a padel coach and give me a list I could reach out to later. Dots came back with three coaches by name, their price per hour, and the courts where they coach.
While you are in a call with your dot, you can also ask it for admin help. Ask it to draft an email, and then tell it whether the draft is good to send or needs another edit first.
Another case I could share was the meeting prep. If I have a meeting in 15 minutes, I ask Dots to research the person and what they do. Dots already knows about me from my connected apps and past chats, so I don’t have to explain the context first. This meeting prep becomes more valuable over time if you already have past data about the person you’re about to meet. It can read your past email, shared Docs/Drive files, meeting transcripts, and proposal decks, so Dots can help you show up more prepared for the meeting.
I can do this while walking or having lunch, by call or chat, because nothing depends on my computer being awake. Dots itself works the same on my phone as on desktop.
Dheeraj said he really liked the call feature. He’d already tried something similar with Meta’s Muse on his wife’s account for travel planning, and he said it handled fine-tuned search requests well. He sees the same pattern in Dots.
Sending Work to Codex
I asked Dots: “Can you deploy three Codex threads to find me new ideas for my newsletter post?” It launched three separate Codex tasks. When I opened one, called “Find emerging AI angles,” you could see the prompt was written by Dots, not me, followed by the results. From there, Dots pulls the results of all three threads into one summary for me.
Editing Alongside Dots in ChatGPT Space
Then I asked Dots to save the ideas into a page. This is ChatGPT Space, a new area that works a lot like Notion. Type a slash and you get headings, bullet lists, images, visualizations, and pages. Inside a page, ChatGPT and your dot can work on the same document with you.
So when I didn’t like one of the ideas, I highlighted it, left a comment, and mentioned Dots, asking it to figure out other ideas because I didn’t like that one. Dots picked the task up from there. I also mentioned ChatGPT in the page and asked it to run another round of research for better ideas, and it started adding text directly to the page.
Slides showed up in my Space during the session. The idea is that you ask for a deck, edit it directly, add slides, insert images, or ask ChatGPT to generate an image and place it on a slide.
Separately, I’d recently asked Dots to turn one of my newsletters into a presentation. It produced both a PowerPoint and a PDF from a simple prompt. The design wasn’t very visual, but the slides were well structured, and I could annotate the parts I wanted changed and let Dots keep editing.
The last thing I showed was ChatGPT Sites. I can mention one of my sites in the Dots chat and ask for a new feature, a file change, or a code review, and Dots handles it.
The way I see it, OpenAI and Anthropic are both building one central place where you work with your agent. You get the output, and you also get enough control to edit and adjust it, because you don’t want the agent to generate everything without you.
If You Live in Claude, Here’s How It Compares
Dheeraj pushed on this, since he’s an Anthropic power user. Claude Code has remote control, cloud sessions, and people leaving a Mac mini running to reach it from their phone. So what does Dots add?
For me, the difference is setup. I mostly run Claude Code locally on my computer. Running it in the cloud, so it doesn’t need my computer awake, still takes extra setup before it works, especially once keys and integrations are involved. Dots only needs connectors. I turn it on and it runs.
Claude’s new projects feature, currently in beta, should close some of that gap. As I covered in my previous post, you run it from your folder, and that folder is copied into the cloud session for you. Then you can work with Claude to read, write, and edit files without requiring your computer to stay awake. I think when this feature is open to everyone, it’ll matter to a lot of people.
The other factor is the model.
I think Opus 5.5 is the smartest model right now, but I still find myself using ChatGPT because it has a better interface, and Astra is good enough that I can live with it for this kind of work. I also like their Sites feature, which is why I keep shifting between Claude and ChatGPT.
Where Dots Still Comes Up Short for Me
Near the end, one of audiences named Vera asked about weaknesses. The first thing I said was that I haven’t used Dots for writing or deep research. I keep it on admin-level work like email, bookings, and simple research.
And even with the nudges, I still end up using Grok Bot more overall. The main reason is that Dots gives you one dot, while Grok Bot lets you build multiple bots with completely different personalities and jobs.
One of my Grok Bots is my padel coach.
The court I play at has an AI camera that sends me a video of my match. I send that video to my padel coach bot, and it analyzes my movement, what I need to improve, and where my weaknesses are. I don’t want to fold that into the same single dot that handles my email and sponsorship deals. OpenAI says it plans to add the option to run more dots, so this may change.
A few other limits I ran into:
You can’t choose the model. Dots always runs on Astra, and I don’t know whether it’s low, medium, high, or ultra. OpenAI keeps those settings blurry, so you don’t really know what’s inside.
Usage is hard to read. In the session, I said Dots seemed to share usage with ChatGPT and Codex. OpenAI’s help page describes a separate allowance for your dot, with extended limits in the first month.
It’s one long conversation. There’s no “new session” button. Everything runs inside one chat, and any compacting happens in the background without telling you. But Grok Bot also works the same way.
It only knows what it can reach. Dheeraj asked about business context scattered across databases with no connector. He summed it up himself: if the data connects through a connector, Dots can use it. Otherwise, you need another way to get it in. My personal skills do come along, though. I have 116 of them, and Dots can follow them when I ask for something.
Where I’d Start If You Have Access
Here’s the simplest rule I gave in the session: if you already use ChatGPT a lot, Dots makes the most sense. If you mostly live in Claude Code, you can probably stick with Claude for now. Your work files are already in ChatGPT, so that’s where Dots has the most to work with.
If you’re on Pro or Business Premium, I’d start small. Connect your email and calendar, then set one morning routine that tells you what needs attention today. I’d give it some normal use before judging it, since the proactive research works by connecting new information to what it already knows. Then see which nudges show up.
If those nudges catch something you hadn’t gotten to yet, like my sponsor reply, you’ll know whether it’s worth adding more.
Dheeraj is still waiting to see Anthropic’s answer to Dots and Grok Bot, and so am I.
Next episode, we’re going through the latest Claude desktop app updates, including what you can now do with Claude Code inside the app without a separate code editor.
Show Details
Show: One Shot Show
Episode: 32
Topic: OpenAI Dots, proactive nudges, and ChatGPT Space
Hosts: Wyndo and Dheeraj
Timestamps
00:35 Episode 32 intro: from OpenClaw to Grok Bot, Muse, and Dots
03:03 Who can use Dots and why always-on agents are the trend
06:28 Proactive nudges and the Grok Bot comparison
07:57 Dots’ cloud computer, connectors, and memory
09:24 Calling your dot
11:38 Scheduled tasks and proactive research
12:51 Dheeraj asks how Dots compares to Claude
18:07 How proactive nudges decide what needs you
19:29 Routines and the chief of staff positioning
21:01 Use cases: morning priorities, voice admin, meeting prep, AI news, Space, Codex, Sites
24:41 Dheeraj on setup, prompting, and scattered business context
27:10 No setup: context from conversation, files, and connectors
30:03 Skills, the model, and usage limits
32:11 Connecting your local computer
33:01 Turning a newsletter into a deck
34:55 Finding a coffee shop near an Airbnb over a call
36:01 Proactive nudges in the live chat
37:17 Finding a padel coach in Bali
38:42 Reading today’s Codex conversations
39:54 The 8 a.m. Codex recap routine
40:54 Overnight updates
41:32 Deploying three Codex threads
43:03 ChatGPT Space and Pages
46:41 Building slides in Space
48:39 Editing ChatGPT Sites from Dots
49:24 Central work hubs and Claude’s simplified app
53:21 Vera’s question about weaknesses, plus the Grok Bot padel coach
58:02 Next episode: Claude desktop app
Resources Mentioned
OpenAI dots: Always-on agent inside ChatGPT with its own cloud computer, proactive research, scheduled tasks, and calls. First dot included on Pro and Business Premium (as of October 2026).
GPT-6 Astra: The OpenAI model Dots runs on.
ChatGPT Space: Shared area for Pages where you can work with ChatGPT and your dot. Slides and Sheets are listed as coming soon on OpenAI’s help page.
ChatGPT Sites: ChatGPT feature for building websites, which Dots can edit.
Codex: OpenAI’s agent for coding and longer tasks. Dots can launch and read Codex threads.
ChatGPT scheduled tasks: Recurring routines Dots runs at set times.
Grok Bot: xAI’s agent product for building multiple named bots with their own cloud computers. Wyndo’s main agent setup. It recently added its own email, so you can send tasks to a bot by email.
Grok 4.7: The xAI model behind Grok Bot.
Meta Muse: Meta’s personal agent. Dheeraj tested it for travel planning.
Instinct: Invite-only personal agent mentioned by Dheeraj. Dheeraj doesn’t have access yet.
OpenClaw: The autonomous agent Dheeraj named as the starting point for this wave of personal agents.
Claude Code: Anthropic’s coding agent, which Wyndo mostly runs locally.
Claude Code remote control and cloud sessions: Ways to reach Claude Code without sitting at your computer, raised by Dheeraj.
Claude projects (beta): Claude’s newer way to run work in cloud sessions, currently in beta.
Claude desktop app: Topic of the next episode. Dheeraj and I noted that Cowork is now merged into the simplified Claude app.
Opus 5.5: The Anthropic model Wyndo considers the smartest right now.
Wispr Flow: Meeting notes source Wyndo uses for client and course context in Dots.
Granola: Meeting notes tool Dots can summarize from.
Notion: Connected app Dots can search for client status updates.
Slack: Connected app and delivery channel for Dots routines.
Email and calendar: The main sources behind Dots’ proactive nudges.
Connectors and MCP: Dheeraj’s summary of how outside business data can reach Dots.
Personal skills: Wyndo’s 116 skills, which Dots can follow.
Blender, FreeCAD, 3D Slicer, Chrome: Apps installed on the dot’s cloud computer.
PowerPoint, PDF, and Google Slides: Formats Dots produced for the newsletter deck, and the tools Space slides could replace.
Airbnb: Booking confirmation email Dots used for the coffee shop and gym search.
AI court camera: The padel court camera that records Wyndo’s matches for his Grok Bot coach.
Mac mini: Dheeraj’s example of an always-on machine for Claude.
