Should you switch from Grok Bot to Dots? What I moved after a day

October 4, 20269 min read

ai-tools / build-notes

A gold line rising across a dark field, cover image for "Should you switch from Grok Bot to Dots? What I moved after a day"

Switch to Dots if you have used ChatGPT for a year or more and prefer one agent. Dots works from that history, while Grok Bot starts from scratch. Stay on Grok Bot if you want a done-for-you setup or several specialized agents, since Dots allows one per paid account. Among the rest, Codex in the cloud covers what Devin sells, Spaces is close to Notion but not a replacement yet, and the Decisions API is a preview to test.

I spent my last few videos recommending Grok Bot. After OpenAI's DevDay I moved a large share of my Grok Bot workflows to Dots within a day, so this is a correction as much as a review. Everything below comes from one day of use. If something breaks next week, I will say so.

Chapters in the video

Who should switch from Grok Bot to Dots

The deciding factor is whether you want a product or want to build on a model. Grok Bot is a product. It is a done-for-you agent setup that needs very little configuration, and you pay heavily in compute and tokens for that ease of use. For a nontechnical person who wants agent workflows without building anything, Grok Bot is still a good choice.

OpenAI is a model company, and Dots sits on top of the model and everything you have already built with it. If you want more advanced workflows, or you want to use the memory, skill files and conversations you have built up in ChatGPT, Dots is the better fit. In my experience the move to Dots is also easier and cheaper to run. I compared the costs in my ChatGPT 6 Astra versus Grok Bot breakdown and free skill file.

The setup cost of switching is real. I spent time configuring Grok Bot the same way I spent time configuring Hermes agents before it, and a better option arrived within a couple of months. That is how AI tooling works right now. Expect to rebuild, and keep your setups light enough that rebuilding is cheap.

Dots replaces a pile of chats with one agent

Dots turns ChatGPT from a list of separate conversations into a single agent that keeps working for you. Until DevDay, ChatGPT was one chat for email, one for a report, one for a landing page. Heavy users have thousands of these spread across ChatGPT and Codex. OpenAI describes Dots as an always-on agent that learns what matters to you and keeps working on your behalf.

Setup is short. You give your Dot a name. I named mine Mad Max. From there you can message it, email it, add it to Slack or call it. I tried calling mine and it hung up on me. Like Grok Bot, Dots gets its own computer in the cloud, so it keeps working after you close your laptop.

The clearest example I have seen comes from the YouTube channel Futurepedia, which had early access before launch. The Futurepedia host sent his Dot one short message: write three intros for an upcoming video, pick the best, turn it into motion graphics and make a thumbnail. Dots wrote the intros in one window and picked one. It handed the winner to Codex, OpenAI's coding tool, which built the animation. While that ran, it opened another chat and worked on the thumbnail. One message produced a script, motion graphics and a thumbnail.

Advanced users could already chain this together in ChatGPT or Claude. What is new is that Dots does it from one message, using context it already has.

Your ChatGPT history is now your biggest advantage

The memory, skill files and conversations you have built in ChatGPT are what make Dots useful on day one. Grok Bot launched about a month ago and had to learn about each user from nothing. Meta's Muse arrived a couple of weeks ago with the same problem. I have used ChatGPT for close to two years, so my Dot had a lot to work with from the start.

Dots runs on ChatGPT 6 Astra, the model OpenAI released about a month before DevDay. A strong model combined with years of personal context is the part Grok Bot cannot match for a long-time ChatGPT user. If you started using AI a year ago and kept at it, you have a head start over someone starting today, because your assistant already knows how you work.

Your job moves from prompting to reviewing

With Dots, the agent brings work to you and you approve or correct it. For the last few years, the people who did well with AI were the ones who prompted well, which mostly meant writing clear instructions. With Hermes or Grok Bot, I still had to tell the agent to look into a project, draft a plan and spin up sub-agents. In Jake's words, "you move from being the instigator to the reviewer."

Within its first day, my Dot found a loose API key that my daily security audit, a cloud automation running on Fable 5.1, had missed. It flagged the issue and proposed the fix without being asked.

Keep that review step in place. Swarms of agents working around the clock tend to produce spam emails, annoyed customers and AI slop. A human in the loop is what stops that.

The one-Dot limit is the main reason to wait

Dots gives you one agent per paid account. Enterprise plans may get a few more, as I understand it. If you like talking to a single agent in one thread, that limit will not bother you. If you rely on several specialized agents, each with its own role, one Dot may feel cramped. In that case, keep Grok Bot running for those workflows until OpenAI changes the limit.

Spaces and Pages are close to Notion, but I am keeping Notion

Spaces is OpenAI's answer to Notion, and it is strong enough that I expect to drop Notion within months, though I have not dropped it yet. A Space is a shared home where your team, ChatGPT and your Dot all work from the same knowledge base. A Page is a document inside a Space, which OpenAI built for people and agents to edit together. If you have used Notion, Pages will feel very familiar.

In a Space, I can tell Mad Max to add a weather forecast to a page and it does. I can also share that page with other people's Dots so they work on the same document.

The client case is where this gets useful for operators. You send a client a report. The client skims it on a phone and asks the page a question. Your Dot answers using the report data, past meeting notes and emails it already has. The same pattern puts pressure on project management tools like Asana and Monday.

Notion is still the best library for my agents to work from, so I am staying for now. Watch how Pages develops before you migrate a knowledge base your team depends on.

Codex in the cloud does what Devin sells

Codex can now keep working in the cloud after you close your laptop, which was the main selling point of both Grok Bot and Devin. Codex also has more context about you than either, because it shares your ChatGPT history. Moving from local to cloud is one switch.

A few months ago, people were buying up Mac minis to run agents locally because nobody trusted agents in the cloud. That workaround is no longer needed for most personal work. I can start a Codex session from the ChatGPT app on my phone and have an app built by the time I get back to my desk.

Trust still has limits. During setup, ChatGPT asked if I wanted to connect my bank accounts to manage subscriptions, and I said no. If you handle client data or are bound by security commitments, keep sensitive work local and test cloud agents slowly.

The Decisions API is a review gate, and still a preview

The Decisions API is OpenAI's answer to Jev, and its value is that the model can only answer from options you define. You give it context, as text or images, plus a fixed set of answers, and it picks one. OpenAI suggests using it to classify content, route requests or choose an agent's next action. Jev, from Typesafe, went viral for the same idea. I have not used Jev, and it is invite only.

For a marketer, the use is lead routing and CRM updates. A lead is qualified or it is not. It belongs in one of your lifecycle stages. It goes to one of your reps. I have had agent workflows in HubSpot create lifecycle stages that did not exist. With the Decisions API, "it can only say what's on your list." When the possible answers are limited, you can check every one. "That's a review gate you can put in front of a client."

The Decisions API is in limited preview, and OpenAI says a broad release is coming soon. Greg Eisenberg, who has a full video on Jev, said he is not sure yet whether OpenAI's version or Jev will win. My recommendation is to test the Decisions API on personal projects and keep it away from client or company systems until it has a track record.

OpenAI cloned Notion and Devin, then partnered with both

On the same day OpenAI released products that compete with Notion and Devin, it launched Sign in with ChatGPT with both companies as partners. Sign in with ChatGPT lets you use the ChatGPT plan you already pay for inside other tools.

My read is that OpenAI wants to own three layers. It wants to be where you log in, where the work lives and where the agent runs. Other companies become a feature inside OpenAI or a partner next to it. "OpenAI wants to be where you log in," and DevDay showed it is ahead in that race.

What to build when your product can be copied in a day

If you are building software, assume someone can give your website URL to ChatGPT 6 Astra and have a working copy within a day or two. Features alone will not protect you. Customer experience will. My background is customer relationship management, and the gap I see is that very few companies think hard about what a human, non-transactional experience looks like. If you can make your customers feel known, and train your agents to support that, you have something a clone cannot copy quickly.

Pick one Grok Bot workflow that depends on context ChatGPT already has about you, rebuild it as a single message to your Dot, and run both side by side for a week before you cancel anything.