Monitor public documents for changes affecting teaching materials
Hello, kazu here. Task: Hello, kazu here. Do you find yourself explaining the background and progress of a task from scratch every time you ask an AI to do something? I think there are times when you feel, "I wish I had an assistant who could remember this work and help me continuously." Today (Japan time), OpenAI announced Dots. Dots is an AI agent aimed at handling such ongoing work. You communicate your goals and the scope of delegation, and it continues to work even between conversations. In addition to answering regular questions, it is a service for thinking about how to work with an AI that has a specific role. In this article, based on official information, I will organize what you can do with Dots, the usage requirements, how to get started, and the initial request prompts you should use. What is OpenAI Dots? Dots is an always-on AI agent powered by GPT-6 Astra. It has its own dedicated cloud computer and carries out ongoing tasks while utilizing connected apps and the context of your conversations. OpenAI suggests starting by creating one dot and giving it a name. While the concept of multiple dots working as a team has been shown, it is best not to think of that as a feature anyone can use from the start. Source: OpenAI "Introducing dots" As an example of using it for content creation, the following interaction is also introduced officially. This is a flow where the AI prepares posts, introductions, and short video drafts based on interview content, and a human provides corrections like, "Keep the photo as is, but shorten the text." The screen also shows that the created draft has not been published yet. For example, if the job is to create educational videos, it can be replaced by handing over a transcript and asking, "Create a draft for the video introduction and announcement. Do not publish it." This is a division of labor where a human reviews the content after completion before publishing. This educational video example is my own suggestion based on the official use cases. Image source: Content creation example from OpenAI's official announcement. This is an official introductory image and not a result of my own actual machine verification. Looking at the official demo screen makes it easier to grasp the relationship between the conversation and the work environment. The left side is the conversation for exchanging requests and progress, and the right side is the computer screen where the work is being done. It shows not just returning answers in text, but also the process of carrying out tasks on the screen. At the bottom, you can also see "Take over" for a human to take control. You can imagine a way of using it where, after requesting work, you check the status in progress and take over if necessary. Image source: Preview image from the demo video in OpenAI's official announcement. This is not a screen I operated myself. If I relate this to the "AI employee" concept I have written about before, it is a story about how to operate an AI that has assigned duties, in addition to being a partner for one-off questions. However, just giving it a name or a job title does not define the scope of work. Communicating what to proceed with and where to check with a human is what leads to actual usability. What can you do with Dots? The official documentation provides examples of use cases such as development, document creation, and content production. For example, preparing revisions based on customer feedback or creating drafts for social media posts from transcripts. Interactions are envisioned to take place not only in ChatGPT but also in Slack, Teams, and other platforms. However, the availability conditions for each feature are not the same, and some text message integrations are currently on a limited waitlist. Source: Official announcement of use cases and communication methods If you want to incorporate these features into your work, it is easier to test them by defining specific roles rather than just saying "help me with everything." The following are examples of how I thought to use them, not actual execution results. Role: Organizing candidates for teaching material revisions I have it read changes in specified official information and organize which parts of the teaching materials are affected. It provides candidates for revisions and the rationale, and a human decides whether to actually reflect them in the materials. Role: Checking drafts before publication; tools: Dots. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include access to OpenAI Dots (Pro/Business Premium) and defined public document sources. Integration uncertainty exists regarding automated monitoring frequency and notification channels. Human approval is strictly required before any changes are applied to teaching materials. Source reliability is secondary, based on the author's proposed use case..
Tools
Published on Agent Work Atlas:
Overview
Hello, kazu here.
Workflow
Hello, kazu here. Do you find yourself explaining the background and progress of a task from scratch every time you ask an AI to do something? I think there are times when you feel, "I wish I had an assistant who could remember this work and help me continuously." Today (Japan time), OpenAI announced Dots. Dots is an AI agent aimed at handling such ongoing work. You communicate your goals and the scope of delegation, and it continues to work even between conversations. In addition to answering regular questions, it is a service for thinking about how to work with an AI that has a specific role. In this article, based on official information, I will organize what you can do with Dots, the usage requirements, how to get started, and the initial request prompts you should use. What is OpenAI Dots? Dots is an always-on AI agent powered by GPT-6 Astra. It has its own dedicated cloud computer and carries out ongoing tasks while utilizing connected apps and the context of your conversations. OpenAI suggests starting by creating one dot and giving it a name. While the concept of multiple dots working as a team has been shown, it is best not to think of that as a feature anyone can use from the start. Source: OpenAI "Introducing dots" As an example of using it for content creation, the following interaction is also introduced officially. This is a flow where the AI prepares posts, introductions, and short video drafts based on interview content, and a human provides corrections like, "Keep the photo as is, but shorten the text." The screen also shows that the created draft has not been published yet. For example, if the job is to create educational videos, it can be replaced by handing over a transcript and asking, "Create a draft for the video introduction and announcement. Do not publish it." This is a division of labor where a human reviews the content after completion before publishing. This educational video example is my own suggestion based on the official use cases. Image source: Content creation example from OpenAI's official announcement. This is an official introductory image and not a result of my own actual machine verification. Looking at the official demo screen makes it easier to grasp the relationship between the conversation and the work environment. The left side is the conversation for exchanging requests and progress, and the right side is the computer screen where the work is being done. It shows not just returning answers in text, but also the process of carrying out tasks on the screen. At the bottom, you can also see "Take over" for a human to take control. You can imagine a way of using it where, after requesting work, you check the status in progress and take over if necessary. Image source: Preview image from the demo video in OpenAI's official announcement. This is not a screen I operated myself. If I relate this to the "AI employee" concept I have written about before, it is a story about how to operate an AI that has assigned duties, in addition to being a partner for one-off questions. However, just giving it a name or a job title does not define the scope of work. Communicating what to proceed with and where to check with a human is what leads to actual usability. What can you do with Dots? The official documentation provides examples of use cases such as development, document creation, and content production. For example, preparing revisions based on customer feedback or creating drafts for social media posts from transcripts. Interactions are envisioned to take place not only in ChatGPT but also in Slack, Teams, and other platforms. However, the availability conditions for each feature are not the same, and some text message integrations are currently on a limited waitlist. Source: Official announcement of use cases and communication methods If you want to incorporate these features into your work, it is easier to test them by defining specific roles rather than just saying "help me with everything." The following are examples of how I thought to use them, not actual execution results. Role: Organizing candidates for teaching material revisions I have it read changes in specified official information and organize which parts of the teaching materials are affected. It provides candidates for revisions and the rationale, and a human decides whether to actually reflect them in the materials. Role: Checking drafts before publication
Reported result
The source does not report measured outcomes or success metrics for this specific workflow. It is presented as a conceptual role definition rather than a verified execution result. A measurable acceptance check would be tracking the number of valid revision suggestions accepted by humans per month versus total suggestions generated.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include access to OpenAI Dots (Pro/Business Premium) and defined public document sources. Integration uncertainty exists regarding automated monitoring frequency and notification channels. Human approval is strictly required before any changes are applied to teaching materials. Source reliability is secondary, based on the author's proposed use case.
Try it
Act as an AI agent using OpenAI Dots to monitor public documents for changes affecting [replaceable: specific teaching material topic]. 1. Input: [replaceable: URL or file path of public document source] and [replaceable: current teaching material content]. 2. Task: Compare the latest public document version against the previous version. Identify any changes that contradict or require updates to the teaching material. 3. Output: Generate a report listing 'What changed', 'Source URL', 'Impact on teaching materials', and 'Suggested revision'. Do not modify the teaching material directly. 4. Human Check: Present the report to me for review. Wait for my confirmation before proceeding with any further actions. 5. Error Handling: If the source document is inaccessible or unchanged, report 'No action required' and stop. First, perform a dry run on one sample document. Show me the output format and accuracy before setting up ongoing monitoring.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.