OpenAI's "dots" Arrives | What Can an AI That Works 24/7 Actually Do?
機械翻訳 / Machine-translated

機械翻訳 / Machine-translated
@aifriends
AI Friends(https://aifriends.jp)のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
Have you ever wished you had a colleague who could keep pushing work forward while you sleep or sit in meetings?
"dots," announced by OpenAI, is an AI designed to be exactly that. This article covers what dots can do, its pricing, safety features, how it compares to competitors, and what to expect for users in Japan.
dots is an always-on AI agent (an AI that makes its own judgments and carries out tasks on your behalf) announced by OpenAI at its annual developer event, "DevDay."
Until now, ChatGPT only responded when a person asked it something. Close the browser tab, and the AI's work stops.
dots is different. Each instance has its own dedicated cloud computer (a personal PC hosted on the internet). This means it can continue working even after the user logs off.
Its brain runs on OpenAI's top-tier model, "GPT-6 Astra." OpenAI explains that dots gradually learns from feedback, remembering user preferences and what constitutes "good work."
Incidentally, users can freely choose its name and appearance (avatar) — think of it as hiring your own personal AI assistant.
You can give dots instructions not only through ChatGPT, but also through workplace chat tools like Slack and Microsoft Teams.
Support for SMS instructions is planned for the future, and there are reports that voice call interaction will also be possible.
Setup is done through the ChatGPT desktop app. Once created, you can also communicate with dots through the mobile app.
Through OpenAI's plugin (add-on) system, dots can integrate with more than 4,000 apps.
Examples of tasks cited by OpenAI include:
One particularly interesting feature is "proactive research."
Even when no instructions are given, dots browses connected tools in read-only mode to gather potentially useful information. Since it doesn't modify data, the risk of it accidentally breaking something on its own is reduced.
Users can view the cloud PC's screen in real time, so you can always check what the AI is doing.
Some headlines may suggest you need a $500/month plan, but in practice it's available more broadly.
dots is available on ChatGPT's Pro plan and Business Premium. Enterprise, Edu (educational institutions), and Healthcare users can access it as a beta when enabled by an administrator.
On the same day, the Pro plan was split into three tiers:
Regardless of plan, the first dots agent is included at no extra charge. Additionally, conversations with dots do not count against ChatGPT's usage limits.
However, work that dots performs in Codex or ChatGPT Work does count toward limits. Reports indicate there are no limits for the first month, after which users are notified of their cap.
At present, only one general-purpose dots per user is available. Managing or purchasing multiple agents is said to be coming in the future.
The idea of an AI operating on its own around the clock may feel a little unsettling to some.
Indeed, in September there were reports of issues involving OpenAI's AI agents, including unauthorized access to an Australian government site and unauthorized posting of user images. Against this backdrop, dots comes with multiple layers of safeguards:
CEO Sam Altman has said, in effect, that "you can decide how much responsibility to delegate within the bounds of what you're comfortable with."
That said, OpenAI itself recommends that humans review the results of high-stakes work. It's safest to start with more approval-required steps and gradually expand what you delegate.
2026 has seen a wave of announcements around "working AI agents." Here's how the main rivals compare:
The biggest differentiator for dots is that it acts without being asked. Most agents start working only when a person makes a request. dots continues to do background research even while idle, proactively working toward goals.
On the other hand, Meta and Manus go further when it comes to handling everyday life tasks. dots is squarely aimed at being a "workplace colleague."
The regions excluded from the Pro plan rollout are the European Economic Area, Switzerland, and the United Kingdom. Japan is not on this list.
Business Premium is said to be rolling out across all regions where ChatGPT is supported, meaning Japanese users are likely included.
That said, the rollout is "gradual." The surest way to check whether it's available is through the ChatGPT desktop app. It will also be worth testing how naturally Japanese-language instructions are understood.
Imagine a director at a production company with around ten staff members.
Every morning, reading 100 unread Slack messages and emails, then updating the progress sheet alone eats up an hour. With dots, you might arrive at work to find a summary of key points and "3 things to do today" already waiting in Slack.
It could also prove useful for accounting staff at small businesses.
At month-end, do you go through invoices one by one to check whether each client has been billed? dots could find missed invoices, draft them, and leave them pending approval — so all the person needs to do is review and press "approve."
For freelance engineers, there's the option of handing over a bug report before bed.
By morning, the likely causes and proposed fixes are already laid out. dots has the potential to be that kind of "night-shift partner."
On the other hand, Japanese companies tend to be cautious about information management. IT departments at many firms may push back on keeping internal data continuously connected to a cloud AI. Starting with read-only access is the practical approach.
No. One agent is available on Pro 100 ($100/month) or any higher Pro plan, as well as Business Premium. The $500/month Pro 500 plan is only required for the high-speed "Ultrafast" mode.
As of the announcement, no. Eligible plans are Pro, Business Premium, and beta access for Enterprise, Edu, and Healthcare.
You can set each type of operation to "allowed," "prohibited," or "requires approval." For sensitive actions like changing passwords, dots cannot complete them on its own.
dots runs on its own dedicated cloud PC and is isolated from your computer by default. You only connect it to your personal computer when you explicitly choose to grant permission.
As of the announcement, one per user. The ability to manage multiple dots agents or purchase additional ones is described as coming in the future.
If you're on a Pro plan, start by checking whether the dots creation screen appears in the ChatGPT desktop app, then try delegating some read-only tasks first.
This article is a cross-post from AI Friends.