I Gave ChatGPT a Job and Walked Away: What ChatGPT Dot Actually Does — and Is It Worth $100 a Month?
The story that opens the video is simple, and it’s the whole pitch for always-on agents in one scene. Anthony gave ChatGPT a job, walked away while at work, and came back to a mediocre result. Before he could even type a correction, a new agent — ChatGPT Dot — had jumped into the thread and given the first agent feedback on its own.
That, he says, is when it clicked: this thing has autonomy. It makes decisions on its own.
Dot is OpenAI’s answer to the personal agents that have been taking over lately, from Grokbot to Muse. It is an always-on agent that lives inside your ChatGPT account, with its own cloud computer and its own browser. You can message it, you can call it, and it is available on desktop, web, and mobile — and it can even be put into Slack and Teams.
The key difference from a normal chat: you can close the chat and it keeps working. It is not tied to your computer being on.
What it costs — and who can get it
At the time of recording, Dot was only available on the $100-and-up monthly plans. It is also available with Business Premium and Enterprise, if your admin enables it.
That frames the whole video’s question: are these capabilities worth increasing your subscription from $20 to $100 a month? What is the price of intelligence like this?
Easy mode: talk to your Dot
In easy mode, Dot is just a chat agent with a personality. Anthony’s is called Sack Boy, a little avatar inspired by the character from LittleBigPlanet.
Because it is baked into ChatGPT, it apparently runs on the newest flagship model, Astra, and you can talk to it like a normal chat and get answers. Ask it what it knows about you and your work, ask for three things it could take off your plate this week, and it recalls from past chats and other context — goals for the quarter, video production prep, an SEO assignment, weekly reporting assembly.
Useful, but as Anthony notes, quick answers are something ChatGPT already does. Easy mode is not the reason to pay more.
Medium mode: let Dot run ChatGPT for you
This is where Dot starts to separate from regular ChatGPT. If you already use ChatGPT with loads of projects and threads in your sidebar, Dot basically needs no onboarding — it already has the context you use every day.
Anthony sends it into his old projects — 10X Media, the content machine, and packaging world — and asks each one for its learnings. With a little guidance (copying the deep link for each thread you want it to work inside, and pasting those links in), Dot spins up three unique threads, one per project, and he can click into any of them at any time to see what is going on, interrupt it, or steer it in a new direction.
That access to all the context you already use every day is, in his view, the thing Muse and Grokbot do not have. It makes onboarding dramatically faster.
A real build: Daybridge, a custom Mac app
The longer test is a build that runs over multiple hours. Anthony asks Dot to recall a prompt he sent the night before: build a robust app on his Mac that combines his Apple Calendar and Reminders in one easy view, do it in Codex, supervise the session, and approve whatever needs to be approved.
Over the course of the build, Dot sends multiple messages — a supervision check, a report on tool availability, the architecture — and steps in when Codex finishes a step and says, in effect, “Okay, done. Approve my next step.”
The output is Daybridge: a view of what is on his calendar today, his wife’s calendar, with reminder lists and calendars that can be toggled and filtered by events, reminders, and more. Anthony loves the UI and the logo and tagline, “a little clarity for your day.” What he would push Dot on for the next revision are the human touches an AI can’t always know — animation, better contrast (the cream and green blend into the background a little), and finding references to take it a step further.
Calling Dot from your phone
Everything so far was at a desk, but the point of an always-on agent is that it is on when you are not there.
Anthony’s example: you go out to lunch and realize you are about to be late for an important meeting. From the mobile or desktop app, you can call your Dot — it actually rings — and ask it to summarize a PDF sitting on your MacBook’s desktop and send that PDF to a coworker, Matt, in a DM.
Dot gets to work the same way it does on desktop: it pulls up the PDF from his Mac, creates the Slack DM, and gives him a link to follow the task. One wrinkle: Codex Mobile is disabled by his workplace admin, so to see that thread he would have to jump back on his work computer.
But Dot also has its own computer, so your computer does not always have to be on. Click on your Dot, open the computers tab, and you will see your connected MacBook alongside Dot’s own virtual machine. The VM can be a little slow at this point, Anthony notes, but asking Dot to reboot it usually gets it back up to speed.
That own computer matters because Dot is not using connectors or plugins or an MCP server in the way Claude or ChatGPT normally do — it is actually clicking around the internet in its own browser. Tools like HubSpot, X, or especially LinkedIn, where accessing analytics data is a pain, become something your little buddy can take care of for you in its own browser, with your login.
God mode: put recurring work on a schedule
Stack all of that up and put it on a repeated schedule, and you get god mode.
Anthony’s use case is the Monday analytics memo. What he used to do was take over the browser himself, click around YouTube Analytics into advanced mode, export the data, paste it into a ChatGPT thread, and ask for analysis — then use the Notion connector to write it up. Multiple steps, waiting around.
Now the prompt is: every Monday at 8 a.m., open YouTube Studio, LinkedIn Analytics, X Analytics, Meta Business Suite, Beehiiv, and GA4 on your computer. Pull the last seven days versus the prior seven days. Add in any screenshots the team sent in Slack in the media operations thread. Then write this week’s memo into the Notion database using last week’s memo as the template. Flag what is working, what is not, and anything off track. Message me when it is done.
The same pattern works for monitoring a website so it proactively knows when leads come in, then spinning up a simple HTML page for each lead — budget, target timeline, current volume — before you even open your inbox. Or for recruiting, where the apps look terrible: have Dot monitor the inbox and spin up a simple page to click through.
Anthony also built Movie Night, a personal recommendation app on his iPhone that looks through his and his wife’s Letterboxd, serves up recommendations based on what they have watched and scored, provides links if something is free or on a service they subscribe to, and — because he has his own Jellyfin server at home — can look at what is on his server and recommend new movies to get. Since they both watch The Godfather and The Godfather Part II, logically they should watch number three. Dot watches Letterboxd every day and updates the app daily.
How to set up your own Dot
If you have access to Dots in ChatGPT, the setup is short:
- Find Dot at the top of the browser, click on it, and give it a name — any name you like — and pick from three categories of avatars. Then hit save.
- Give it a goal. Point it toward your projects by giving it those deep links: right-click a thread, copy the deep link, and paste it in for each project you want it to work inside.
- Connect it to your apps in its personal computer, up front, because it is annoying when you get blocked in the moment. Ask it to log into whatever apps and tools you will want it to use regularly — Amazon, YouTube Analytics, GA, whatever.
- Leave your computer open, make sure it is linked, and then just walk away. You can call it, you can text it, and it will execute on the fly.
The verdict in the video
Anthony’s verdict is measured. When Dot is managing threads, it is really good at executing on builds. He is not loving the creative use cases — it still needs more human babysitting to get a lot of things across the finish line.
He is leaning that it is not worth the jump from the $20 to $100 subscription just for Dot. If you absolutely need to try an always-on agent, he recommends going with a free one like Muse or a $20 one like Grokbot. The feature alone is not worth it unless you are already planning on upgrading. It is a nice to have. If you are on a team or enterprise plan, he says to ping your admin to see if you can get access — it is worth a try.
When you start getting messages from your Dot throughout the day, when it is executing work for you and reminding you of important tasks, he says, you will start feeling its power, and you are going to want to keep it around. It is not really like a new employee. It is more like your personal work buddy.
The open question he leaves, for workers, freelancers, and people using agents for their personal life, is what value we place on having an AI thread that sparks itself up continuously. Is it a hundred bucks a month? Will it become more than that? That, he says, is for you to decide.
My take
I don’t buy an agent on its avatar or its demo. I buy it on one test: does work move while I’m away, and can I trust what it did when I come back?
On that test, Dot’s real edge isn’t chat answers — every good model does those now. It’s context plus continuity: it already lives where your projects live, it can run threads in parallel, and it can be put on a schedule so Monday’s memo, or a lead page, or a daily app update just happens. That’s the part that actually takes something off your plate.
The weak spot in the video rings true to me, too. Execution is the strong suit; creative work still needs a human in the loop to push it past “good enough” — better contrast, better references, a sharper angle. An agent that needs babysitting on taste isn’t a replacement for taste. It’s a very fast junior who still needs direction.
And on price, I’m with Anthony for most solo operators: if you’re already on the $100 tier or your admin can turn it on, try it. If you’re jumping from $20 to $100 just for Dot, that jump has to be paid for by recurring work it actually owns — not by the novelty of a ringing phone call. The agent you actually use every day beats the agent with the best demo.
The question that matters isn’t whether an always-on thread is cool. It’s whether you have work worth handing it on a schedule, with clear approval points, so it earns that hundred dollars instead of just spending it.
Would YouRather?
So here’s the real test, and it’s a would-you-rather: would you rather pay $100 a month for an always-on agent that keeps working after you walk away, or stick with a free agent and do the creative steering yourself?
There’s no wrong answer — it depends on how much recurring work you actually have to hand over, and how much you trust an agent to run with it unsupervised. But it’s worth answering honestly, because that answer tells you whether Dot, Muse, Grokbot, or none of them is the right fit for you right now.
Want to see where you land against everyone else? Head over to Normie — where average is awesome — take today’s poll, vote, and find out what the crowd actually thinks. Then come back and tell me: are you team always-on, or team human-in-the-loop
Recent Comments