AI that works even after the conversation ends... OpenAI's 'Dots' joins the personal assistant competition

▲ Photo: AI-generated image

An AI has emerged that remembers ongoing tasks and prepares for the next without needing new instructions every time. OpenAI's 'Dots', unveiled on the 29th, is an always-on AI agent that continues to work towards the user's goals even between conversations. The extent to which it can handle repetitive tasks like checking schedules or organizing data is a matter of interest.

According to OpenAI, Dots operates on its own cloud computer based on 'GPT-6 Astra' and utilizes apps connected by the user. The company explained that it can connect to over 4,000 apps through its plugin ecosystem. However, the number of connectable apps does not mean that it can perform all tasks in all apps. The actual scope of use varies depending on the connected services and the permissions granted.

Practical application in daily life can start with relatively small tasks. OpenAI's official help documentation gives examples such as reviewing schedules, researching topics, and setting reminders. Repetitive tasks, like checking the calendar every morning to inform about the day's schedule, can also be set up. If utilized this way, it would be appropriate to set both the desired outcome and the checking frequency, such as 'summarize today's schedule and organize the materials needed for preparation.'

For freelancers and small business owners, the feature of continuing task preparation stands out. OpenAI introduced a case where an early user's Dot discovered missing billing tasks, prepared invoices, and sent them after approval. It also suggested a use case where a content creator prepares an introduction and social media post drafts based on interview records. These are examples and feature descriptions disclosed by the company, and do not guarantee the same results for all tasks.

Initially, the service is available to some paid plans. According to the official help documentation, Personal Pro will be progressively rolled out to markets outside the European Economic Area, Switzerland, and the UK, while Business Premium will be available in ChatGPT-supported regions. Enterprise accounts require administrators to enable beta features. Free and Plus plans are not included in this initial offering. There is also a notice that it may take several days for the feature to appear in an account.

There is no separate additional charge for the first Dot in Pro and Business Premium, but the workload is not unlimited. Usage for advanced tasks is allocated within the plan. Initial creation is done via desktop app or desktop web, and Dots cannot currently be created on mobile.

The terms 'always-on' and 'unlimited authority' must also be distinguished. OpenAI explains that it uses read-only tools for proactive research, examining useful information without new requests. These research tools cannot send messages to others, change the content of connected apps, or manipulate computers. Subsequent executions must follow separate behavioral rules and safety checks.

When actually using it, it is important to define the boundary between drafting and external execution. For example, instructing 'prepare a draft reply but send it after confirmation' clearly defines the scope of responsibility. Official guidance states that sensitive tasks such as changing passwords or transferring money must be completed by the user directly. It also explicitly states that Dots can make mistakes.

Disconnecting an app is also different from deleting information that has already been conveyed. According to the official help documentation, disconnecting stops new access but does not delete information that the Dot has already remembered. It is easier to manage by determining the materials and services needed for the initial tasks rather than connecting multiple accounts at once.

The practicality of Dots needs to be judged not only by how many tasks it can be assigned but also by how much time it takes to review and correct the results. A realistic approach is to start with tasks with clear scopes, such as schedule summaries or data research, and then examine the accuracy and the burden of verification. While AI continues to work, humans will be responsible for setting goals and execution scopes, and judging important results.