[Government AI GENAI] Agent Chat Elementary
Table of Contents
- Introduction: Agent Chat - Getting Started - How to Use Agent Chat in Your Daily Work
- Case 1: Preparation of draft leaflets and pamphlets for residents
- Case (ii) Creation of a one page summary for an executive explanation based on anticipated questions and answers
- Case (iii) Creating the first draft of the guidelines for holding advisory councils and committees
- Case (iv) Aggregation and trend analysis of the user questionnaire by question
- Summary:
Introduction: Agent Chat - Getting Started - How to Use Agent Chat in Your Daily Work
"[Government AI GENAI] Introduction of Agent Chat" introduced basic operations. In the beginner's edition, we will look at how it can be used in actual operations with specific examples.
The basic use of Agent Chat is the same for any job. Simply by placing materials and date on My Drive and instructing the AI, you can perform a series of tasks such as reading, organizing, and outputting files. Furthermore, by saving work procedures as skills, you can perform the same task repeatedly.
In this article, we will take a look at four common cases in the daily work of the administration, from creating draft versions of information leaflets to aggregating and analyzing questionnaire results. If you leave routine tasks that take time to complete to AI agents, staff can focus on tasks that only humans can do, such as making decisions and adjustments. Please refer to them as you read along.
Case 1: Preparation of draft leaflets and pamphlets for residents
The first case is making a draft of a guidance leaflet for residents.
It is necessary to write administrative notices clearly and accurately. If the expression is too rigid, it will not be read. On the other hand, if it is too informal, the accuracy as an administrative document will be lost. Therefore, the balance between the two is important.
[Operation Flow]
Open Agent Chat and enter the following in the entry field: "Create a Word file for the leaflet for the smartphone course for residents. For the leaflet, refer to the project plan saved in My Drive."
For information that has already been decided, such as date, place, target person, and belongings, it is sufficient to indicate it at the prompt or pass it as a file. If you want to browse to a file in My Drive, select the check box.
After you submit your request, you will see on your screen how the AI Agent is processing it. When processing is complete, the flyer text, consisting of the headline, text, and information on what to bring, will be saved as a Word file in My Drive.
When you open the file, the headings communicate at a glance, and the body text is organized into short, easy-to-read paragraphs. If you want to make a change, simply follow the instructions in the chat. "Change the wording to make it more understandable," I say, and the AI agent will save the revised version to My Drive.
As shown in the following screen, the information has been organized in the revised version, and explanations have been added to each item, making it easier to understand.
[Notes]
When you're focused on making stylistic adjustments, it's easy to neglect fact checking. Be sure to check verbatim for factual information such as the date, time, venue, and audience for the event, both generated and stylistically adjusted. Remember to reread through the final version to make sure that the information you originally entered has not changed, especially if you have made multiple revisions.
Case (ii) Creation of a one page summary for an executive explanation based on anticipated questions and answers
The second case is an example of creating a one page document for executives' explanation from expected questions and answers.
While the anticipated questions and answers are detailed, the time to explain them to executives is limited. Since it is difficult to get the entire text read, it is another task to remake it into a document that can convey the main points in a short time.
[Operation Flow]
First, upload the completed list of anticipated questions and answers to My Drive. This time, we will use a document with a total of 6 pages of questions and answers.
There is no need to rewrite the anticipated Q & A itself, and you can use the materials at hand as they are. All you have to do is ask the agent chat, "Summarize this anticipated Q & A so that it fits exactly on one A4 page. List the important questions at the top, and add the original question number at the end of each item."
After sending the request, you see on the screen how the AI Agent proceeds.
When the process is complete, a one page summary document of key points from the "what if" questions is saved on My Drive. It is organized in a format that is quick for executives to read, and is sorted so that the most important questions are at the top. Note also that the original question number is included with each line item.
Even if there are additional questions from executives who have seen this material, they can immediately refer to the relevant part of the original anticipated questions and answers list from the question number. It is also a feature that the materials can be remade to suit the reader rather than handing over detailed materials as they are.
[Notes]
It is helpful that the questions are sorted so that the questions with the highest importance are at the top of the list. However, the order in which the questions are presented and the content of the verbal support are final decisions at the discretion of the person in charge.
In addition, please be sure to compare both the generated one page document and the content of the original anticipated questions and answers before approval.
Case (iii) Creating the first draft of the guidelines for holding advisory councils and committees
The third example is the creation of guidelines for holding advisory councils and committees.
The outline has certain components. Items such as the purpose of holding the outline, the composition of the committee members, and the method of convening are almost the same in all the outlines. Therefore, it is more efficient to refer to past outlines than to write from scratch.
[Operation Flow]
First, upload to My Drive the guidelines for holding advisory councils and committees that will be used as a reference. This time I prepared 2 kinds.
Ask the agent chat, "Based on the structure of the two outlines uploaded to My Drive, create the outline of the establishment of the regional Data Coordination Platform review committee," and then enter the name of the new committee, the purpose of its establishment, the main points to be considered, and the expected composition of the committee members.
By giving an outline of the new committee, the first draft will be more in line with the actual situation.
After submitting the request, you will see on the screen how the AI agent is proceeding with the process. When the process is complete, a first draft that organizes the necessary items for the outline, such as the purpose of its establishment, the composition of its members, and the method of calling meetings, is saved on My Drive.
In the first draft, please check if the terms of office and dismissal of the committee members follow the writing style of the guidelines.
[Additional instructions]
If you want to further enrich the generated first draft, you can also give additional instructions in the chat.
This time, he ordered to add a provision on the creation of a working group to consider more specialized content.
As a result, the provision for the establishment of a working group was added as a separate article and renumbered as indicated. Documents such as outlines, which have few opportunities to be reviewed once they have been established, can be completed into a highly complete outline by repeating such exchanges at the stage of the first draft.
[Notes]
The more typed a document is, the more effective it is to create a first draft through agent chat. However, depending on the nature of the document and the issue, it may be necessary to check the consistency with the laws and regulations and official directives. Please use the generated document as a first draft and be sure to review it yourself before approval. If you have an existing template at hand, first place it on My Drive and try it out.
Case (iv) Aggregation and trend analysis of the user questionnaire by question
The fourth example is the aggregation of user questionnaires.
For satisfaction surveys conducted after an event or after the provision of services, the more responses there are, the more time and effort it takes to tally them up. Counting the number of cases for each question, summarizing the tendency of free writing, and reporting the results in a table take more time than expected.
[Operation Flow]
What we use this time is a CSV file of response data output from the questionnaire system. The response ID, answers to each question, and the date and time of the response are recorded in one line per case, and the data amounts to 1,000.
Upload this CSV file to My Drive and ask them to "Add up the number of cases for each question, analyze the trend, and summarize it in Excel." Whether you have a few dozen or a thousand or more, you can use the same procedures regardless of size.
After submitting your request, you will see on your screen how the AI agent is working through it. Once the process is complete, an Excel file is saved to your My Drive that automatically tallies the 1,000 responses, question by question, and summarizes the distribution of satisfaction, top remediation requests, and free-text trends. The spreadsheet has been formatted in a way that makes it easy to use as material for report materials.
[Notes]
However, classification and aggregation based on AI may contain errors. Please check the contents against the original data before using them.
[Save and reuse as skills]
These steps can be saved as "skills" and reused.
After a task is completed in Agent Chat, you can say, "Save these steps as skills." The AI Agent creates a markdown file of these steps and writes it to the "skills" folder on My Drive.
This file records the work plan, such as which file was processed in what way, and from what viewpoint it was confirmed and organized.
The next time you perform the same type of task, simply request it using this skill, and the process will proceed as it did before. There will be no need to explain the procedure from scratch, and it will also be helpful when there is a change in personnel.
The more data collected, the more effective Agent Chat is at compiling questionnaires. The time previously spent on manual compilation and analysis can now be used for more value-added work, such as deciphering implications from the compilation results and considering future policies.
Summary:
This time, we introduced four examples. From creating new sentences such as leaflet drafts to compiling and analyzing data, the basic flow of placing materials on My Drive and instructing the work remains the same.
Agent Chat is especially useful for tasks that you find "time-consuming" in your daily work. Try it with everyday tasks first.
"Intermediate Agent Chat" explains how to use the system for more practical purposes and points to note in using the system. Please refer to it as well.
Please refer to the following links for the introduction and intermediate editions of Agent Chat.