1. It's Dot, not Doc, and it acts more like a teammate than a chat window
The name trips people up at first. It's not Doc. It's Dot.
Regular ChatGPT works like this: you ask, it thinks, it answers, and that's it. A Dot works differently. OpenAI calls Dots "always-on agents." They run on GPT-6 Astra, get their own cloud computer and browser, and can use your connected apps to keep working toward a goal over time.
So a Dot is less "the thing that answers your questions" and more "the person who owns this job."
AI just got moved out of the chat box and onto the team roster.
2. "24/7" doesn't mean it's spinning nonstop. It means the job doesn't end when the chat does
This is easy to get wrong. Running 24 hours a day doesn't mean you've got a free server burning CPU every second.
What matters is that the role and the context stick around after the conversation ends. A Dot can keep researching in the background, run regular checks, look at your connected apps, and carry on with tasks that are still in progress. Even with no new message from you, it can read the sources you've allowed and look for things that might help. OpenAI calls this "proactive research."
Before, you had to remember the task and go call the AI. With a Dot, someone already has the job on their desk.
That sounds small, but it's huge. A lot of real work time goes not into the work itself but into figuring out "wait, where did we leave off?"
3. You set it up on a computer, then take it with you on your phone
You create your first Dot in the ChatGPT desktop app or in a browser on your computer. You give it a name and a goal, connect the apps it needs, and decide what it's allowed to do on its own.
After that, you can chat with it from the mobile app. Think of it as registering the birth on your computer once, then carrying it around on your Android phone or iPhone.
Its profile shows what's in progress, what's Scheduled, and what's Completed, and you can change how often recurring tasks run and how you get notified. Letting it use your own computer is optional. By default, the Dot has its own separate cloud computer.
So no, you don't have to leave your PC on all night to keep it running. That part really matters.
4. The usage deal is great, but nobody said "free and unlimited forever"
At launch, your first Dot comes with eligible Pro or Business Premium plans at no extra cost. On Enterprise plans it's a beta that an admin has to turn on.
OpenAI says chatting with a Dot doesn't count toward your normal ChatGPT usage limits. "Deeper work" has its own separate allowance, and that limit is raised for the first 1 month after launch. The release notes also say that for that first month, Dots usage won't count toward eligible users' regular plan allowance.
People on social media have posted things like "I left Blender running overnight and my usage meter barely moved." That fits the launch terms. But jumping from there to "I now have a free, unlimited GPU farm forever" is a stretch.
Also, if a Dot starts or manages a task in Codex or ChatGPT Work, that Codex or Work usage gets counted as usual.
It's not a free lunch. It is, though, a seriously generous free appetizer.
5. The best use might be as a helper for your scheduled tasks
Scheduled Tasks are great at one thing: starting work on time. But in real-world automation, finishing is the hard part.
An API fails once. A permission is missing. Something times out halfway. Just one step breaks. A log gets written. The cause gets labeled. Even the "next step" gets written down.
And then nobody does the next step.
If you've ever automated anything, you know the feeling.
Put a Dot in the loop and the roles split cleanly.
The scheduled task is the one who starts the job on time. The Dot is the one who checks the result and picks it up if it isn't finished.
When it sees a failure, a job that only half finished, a job that's on hold, or something the task couldn't handle, it re-checks where things stand and picks up from the last point that was confirmed done. If there's other work it can safely keep doing, it doesn't stop. If the problem is in the code, it fixes it, hands it to a specialist coding agent if needed, and then collects that result too.
You go from "send a failure alert and walk away" to "someone keeps owning the failed job."
6. Think of it as foreman, clock, and repair crew
Splitting it into three layers makes it easy to design.
Scheduled Task is the clock. It kicks off work at a set time or when a condition is met.
Dot is the foreman. It keeps track of things across runs, picks up what's unfinished, decides what comes first, goes after the root cause instead of just hitting retry, and follows the job until it's truly done.
Codex is the repair crew. You call it in when you need code changes, tests, refactoring, or anything complex to build.
The key point: a Dot shouldn't feel like it's done its job the moment it creates a Codex task.
Start Codex → fix is ready → review the changes → run the tests → ship it → run the job again → check it in production.
Only when that whole loop closes is the job actually done.
The foreman doesn't get to go home just because he called the repair guy.
7. Don't stop at "report the failure." Close the loop
The most important part of what you tell your always-on helper is when the job counts as finished.
A bad instruction: "Let me know if something fails." That just gets you a really fancy security camera.
A good instruction: "When something fails, check the current state, find the cause, make the smallest safe fix, check again, and keep going if anything is left. If the code needs changing, hand it to the coding agent, collect the result, and check it again. If you can't finish right now, save the last confirmed point and the next step, and pick up from there next time."
And if the same failure keeps coming back, go back and rethink the original scheduled task itself: its instructions, when it should stop, what it looks at, and how it decides it's done.
"I've identified the cause" is the work equivalent of "We now know what kind of fire this is."
Great. Now put it out.
8. You don't have to hand your Dot the keys to the whole kingdom
Dots come with permission settings and Custom Rules. You can decide what it can just go ahead and do, what needs your OK first, what it should ask about every time, and what it should hand back to a human.
Proactive research is also limited on purpose. In the background it can read your connected sources, but those research tools can't send messages, change content through plugins, or take over your browser or computer on their own. Any real change goes back through the normal permissions, approvals, and safety checks.
That's less a weakness and more a smart design choice for an always-on AI. Something that works 24 hours a day doesn't also need the power to break your live site 24 hours a day.
A good rule of thumb: let it look and research widely, but keep anything destructive or with big outside impact on a short leash.
9. In real use, "thoughtful" means the answer is waiting before you ask
Once you actually start using a Dot, you find jobs it's great at beyond helping with scheduled tasks.
For example, it can check your traffic regularly and chart how the articles that are taking off are trending. And it doesn't stop at "which articles grew." It also sorts out what those articles have in common: topic, how people found them, language, how they did after going live, where readers clicked next. Then it points out what to improve next before you even ask.
The other one is keeping things running. If something is stuck on the Cloudflare or GitHub side, it checks the state, narrows down the cause, fixes it if that's safe, and follows it through to the end if needed. You don't have to keep asking "Where is it stuck?", "How's traffic?", or "Which articles are growing?" By the time you open the Dot, the digging and fixing are already underway and the answer is waiting.
Quietly, that might be the best part.
And in practice it went beyond just watching traffic. It pushed Cloudflare builds forward, kept an eye on GitHub, and made the fixes that were needed. So it's not a watchdog that just reports problems. Within what you've allowed, it rolls up its sleeves and does the work.
You stop going to the AI for an answer every time. You walk over to your teammate's desk and the chart, the diagnosis, and the fix are already sitting there.
"Thoughtful AI" sounds vague, but in practice it's very concrete: it picks up the questions you used to have to remember and ask every time, and handles them first.
At that point, the real win may not be how fast the AI answers, but how much less "oh right, I need to check that" you have to carry around.
It starts to feel less like giving an AI work and more like dropping by to see what a teammate who's already on it has done.
10. Bottom line: what the elves need isn't magic, it's a clear finish line
The real value of Dots isn't "one more smart chatbot."
It's that the state of a job stays put when you step away, works across scheduled tasks and connected apps, and lets work continue from the last failure.
So it really shines not as someone to chat with, but as a full-time helper for recurring jobs that keep failing, a supervisor who picks up unfinished work, and an operator who calls in a specialist only when code needs fixing.
It's the AI version of "The Elves and the Shoemaker."
But it's not enough for the shoes to be finished by morning. Someone should check that they fit.
The motto for the age of always-on agents:
Don't just report it. Finish it.
