Open AI dots.
You shut your laptop at 6 PM.
The project launch is tomorrow, and the final draft still has three unresolved comments.
You leave it.
For years, the digital tools you rely on have gone quiet the moment you step away.
You ask a question, get an answer, and close the tab. The machine waits in frozen animation until you return.
OpenAI is trying to break that rhythm.
Their new product, Dots, introduces a different relationship with your tools.
A Dot is an AI agent that stays awake.
It remembers what it is responsible for, keeps its eyes open, and continues working between conversations.
A standard chatbot can help you draft a launch email. A Dot could watch the launch unfold for a week.
It could track incoming feedback, revise the supporting material based on customer confusion, and quietly ask for your final approval when a change is needed.
The tool is literally trying to be a colleague.
The difference is responsibility
OpenAI describes Dots as always-on agents powered by the GPT-6 Astra model.
Each one is equipped with a cloud computer, a browser, persistent memory, and access to the applications its owner chooses to connect.
This setup allows the agent to make progress on several ongoing projects while the human is elsewhere.
This represents a massive shift from ChatGPT.
The interesting feature here is continuity. When you give a Dot an ongoing goal, it can independently decide what needs attention next.
It performs permitted work in the background and returns with results. If it reaches a decision that requires human judgment, it pauses and asks.
A software developer could have a Dot watching a stream of customer feedback, identifying minor bugs, testing potential changes, and preparing pull requests for review.
A research scientist could rely on a Dot to rerun a complex analysis the moment fresh data arrives.
These examples are polished launch material. The much harder question is whether a Dot can reliably handle the dull middle of a job.
Real work is full of scattered updates, changed requirements, half-finished documents, and minor decisions that consume a surprising amount of a person's week.
A computer of its own
To do any of this, a Dot works inside an isolated cloud computer. It has the freedom to browse websites, run tools, analyze information, and create files in its own environment. OpenAI says its plugin ecosystem can connect Dots to more than 4,000 different applications. People can reach their Dot through the familiar ChatGPT interface, Slack, and Microsoft Teams.
You can also choose to connect a personal computer, though that access starts turned off by default. If enabled, the Dot can utilize local files and tools, or create work using Codex and ChatGPT Work.
This local connection is a meaningful escalation in trust. A helpful research assistant armed with a disposable cloud browser is one thing. An assistant that has permission to touch the files sitting on your physical laptop is something else entirely.
This architecture is where Dots begins to look less like a standalone model release and more like a new operating layer for OpenAI's existing products. ChatGPT handles the conversational relationship.
Astra supplies the raw reasoning power.
Plugins provide the necessary outside context.
Codex and Work agents perform the specialist jobs.
The Dot sits above all of them, remembers the larger objective, and delegates tasks accordingly.
What a Dot could do on an ordinary Tuesday
Imagine you are preparing a major product launch.
In the past, a normal chat session might help you write the initial announcement text.
A Dot could handle the messy reality of the week leading up to it.
It could check the project calendar first thing in the morning.
It might notice that a critical feature delivery has slipped by two days.
Without being asked, it could compare that schedule change with the draft announcement, prepare a revised version removing the delayed feature, and post the proposed wording for your approval instead of publishing it immediately.
Later in the day, it could collect new customer comments from early testers, group the repeated complaints into themes, and send a small coding task to Codex to address the most glaring UI issue.
By the time you log off, you might have an updated launch document and a tested code fix waiting for your review.
Dots is a massive bet that coordination and continuity are becoming more important than any single isolated model response.
Autonomy comes with a permissions system
OpenAI seems highly aware of this risk and has built Dots around rules rather than unlimited access.
Users have granular control.
They can tell a Dot to act automatically on certain tasks, act only when pre-approved, ask before taking any action, or simply hand a complicated task back to the user entirely.
The company's safety documentation details some firm boundaries. Purchases always require explicit human approval.
Password changes and transfers between financial accounts must be handled directly by the user.
Proactive research tasks use read-only tools by default, ensuring that an unsolicited background task cannot directly alter connected content without permission.
Those controls are sensible, but they will not turn Dots into a flawless digital employee on day one.
Early users will need to treat permission settings as a core part of the job description, not just as annoying setup trivia to click past.
The prompt is no longer the product
Dots is currently rolling out gradually to ChatGPT Pro users outside the EEA, Switzerland, and the UK. Business Premium, Enterprise, Edu, and Healthcare workspaces have managed access in supported regions.
This shifts the entire paradigm of how we judge AI. Dots will be judged by a standard that chatbots have mostly managed to avoid.
A clever text generation answer can be impressive even when it is occasionally factually wrong.
An autonomous assistant with ongoing access to your life has to be dependable, predictable, and easy to interrupt when things go sideways. Crucially, it needs to know when a decision truly belongs to its owner.
I suspect the first genuinely useful Dots will have very narrow responsibilities and boring, restricted permissions.
One might simply watch a shared calendar and alert you to conflicts.
Another might just keep a central project document current based on Slack chatter.
A third might follow customer bug reports but have zero ability to contact anyone.
The cinematic version of an AI that runs your entire life can wait.
OpenAI is selling something far more ambitious than better prompting.
They are selling the ability to hand a piece of software a chunk of real responsibility and then walk away.
If Dots actually works as advertised, we may soon spend far less time asking machines for answers and far more time deciding exactly which parts of our lives they are allowed to manage.
In case we are meeting for the first time, come over here, it'll be worth the roller coaster of articles that are gonna come up in the next few weeks.
If you're an established writer, here are the brands paying for sponsored articles.