Work support device

Abstract

To provide a work support device that can appropriately execute tasks according to the event. [Solution] The memory unit 104 stores a procedure table 300 which describes the procedures for corresponding tasks in response to an event, and an operation instruction table 600 which shows agent operation instructions for each corresponding task, which are support information that assists the operation of agent 210. When the CPU 101 references a work procedure in the procedure table 300 in a task based on an instruction statement, it provides agent 210 with an agent operation instruction corresponding to the target task, which is a corresponding task to be performed according to that work procedure, and causes agent 210 to execute the task based on the agent operation instruction.

Description

Technical Field

[0001] The present disclosure relates to a work support device.

Background Art

[0002] With the spread of IT (Information Technology) services, the importance of troubleshooting IT systems that provide IT services has been increasing. In troubleshooting an IT system, when an event such as an abnormality related to the IT system occurs, an operation manager in charge of the operation and maintenance of the IT system needs to perform appropriate work according to the event. Therefore, when an event occurs, it is common for the person in charge of operation to check a pre-prepared procedure manual and perform work according to the procedure described in the procedure manual.

[0003] However, since events related to IT systems are diverse and their content and occurrence situations are complex, it is not easy for an operation manager to check the procedure manual, identify an appropriate procedure, and perform work according to the procedure.

[0004] In recent years, generative AI (Artificial Intelligence) agent technology that autonomously executes tasks using a generative AI model has been attracting attention. The generative AI model is suitable for generating content according to an input instruction sentence, but is not suitable for executing complex processes such as a process of making a decision and a process including a plurality of steps. For this reason, in the generative AI agent technology, a program called an AI agent compensates for processes that the generative AI model is not good at, and dynamically makes a decision using the generative AI model while interacting with the user. For example, Patent Document 1 discloses a technology capable of classifying the intention of a sentence input from a user.

[0005] By applying this type of generative AI agent technology to troubleshooting IT systems, the AI ​​agent can refer to procedure manuals and dynamically generate information related to appropriate work procedures. This makes it possible to streamline the work of system administrators and expedite tasks in response to events. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] U.S. Patent No. 10740566 [Overview of the project] [Problems that the invention aims to solve]

[0007] AI agents typically consist of multiple agents, each possessing several tools built using Python functions or similar technologies. In generative AI agent technology, these agents collaborate to compensate for processing that the generative AI model is not proficient at. However, simply following a procedure to ensure the appropriate AI agent executes the appropriate tools is not easy.

[0008] The purpose of this disclosure is to provide a work support device that can appropriately perform tasks in response to events. [Means for solving the problem]

[0009] A work support device according to one aspect of this disclosure is a work support device that supports said work using an AI agent that performs a task based on instructions related to said work in response to an event, The system includes a storage unit that stores procedure information describing the steps of each of the aforementioned tasks, and auxiliary information indicating support information for each of the aforementioned tasks to assist the operation of the AI ​​agent, and a control unit that, when the steps in the procedure information are referenced in the task, provides the AI ​​agent with support information corresponding to the target task which is the task performed according to the said steps, and causes the AI ​​agent to execute the task based on the support information. [Effects of the Invention]

[0010] According to the present invention, it becomes possible to perform tasks appropriately and quickly in response to events. [Brief explanation of the drawing]

[0011] [Figure 1] This figure shows the configuration of a management computer according to one embodiment of the present disclosure. [Figure 2] This figure shows an example of a procedure manual table. [Figure 3] This figure shows an example of an operation instruction index table. [Figure 4] This figure shows an example of a group-based operation instruction index table. [Figure 5] This figure shows an example of an operation instruction table. [Figure 6] This figure shows an example of an agent table. [Figure 7] This diagram illustrates an example of an LLM service table. [Figure 8] This is a diagram illustrating an example of task execution processing. [Figure 9] This flowchart illustrates a more detailed example of the task execution process. [Figure 10] This flowchart illustrates other, more detailed examples of task execution processes. [Figure 11] This is a flowchart illustrating an example of the creation process. [Figure 12] This is a flowchart illustrating another example of the creation process.

Embodiments for Carrying Out the Invention

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following description, although the "program" may be used as the subject to describe the processing, the program is executed by a processor (for example, a CPU (Central Processing Unit)) to perform the defined processing while appropriately using a storage unit (for example, a memory) and / or an interface device (for example, a communication port), etc. Therefore, the subject of the processing may be the processor (or a device or system having the processor). Further, the processor may include a hardware circuit that performs part or all of the processing. The program may be installed from a program source into a device such as a computer. The program source may be, for example, a program distribution server or a computer-readable storage medium. Also, in the following description, two or more programs may be realized as one program, or one program may be realized as two or more programs.

[0013] FIG. 1 is a diagram showing the configuration of a management computer according to an embodiment of the present disclosure. The management computer 100 shown in FIG. 1 is mutually connected communicably with an IT system 120 and an LLM server 150 via a LAN (Local Area Network) 110.

[0014] The IT system 120 is a system to be managed by the management computer 100, and includes a plurality of clusters 130 and a plurality of storages 140. The cluster 130 has a plurality of nodes 131 which are computers that perform various processes, and the plurality of nodes 131 cooperate to operate like one system. The storage 140 stores various data generated or used in the processing of the cluster 130.

[0015] The LLM server 150 has an LLM (Large Language Models), and is a generative AI server that performs various processes using the LLM 151. In this embodiment, the LLM server 150 is an on-premises server, but it may be arranged on a public cloud (not shown). The LLM 151 is a generative AI model that generates and outputs content according to the input prompt. In this embodiment, the content is text (character information).

[0016] The management computer 100 is a device for supporting the operation and management of the IT system 120. More specifically, when an event such as an abnormality related to the IT system 120 occurs, it is a work support device that supports the user, who is an operator of the IT system 120, in performing work corresponding to the event.

[0017] The management computer 100 has a CPU 101, a memory 102, a NW (Network) interface 103, and a storage unit 104. The management computer 100 is also connected to a display 105 that displays various information.

[0018] The CPU 101 is a processor that functions as a control unit for controlling the management computer 100. It reads the program stored in the storage unit 104 and executes the read program using the memory 102 as a work area to perform various processes.

[0019] The NW interface 103 is a communication unit that is communicably connected to the IT system 120 and the LLM server 150, which are external devices, via the LAN 110, and performs data transmission and reception with the IT system 120 and the LLM server 150.

[0020] The storage unit 104 is, for example, an HDD (Hard Disk Drive) and stores a program that defines the operation of the CPU 101, and various data generated or used by that program. In this embodiment, the storage unit 104 stores the operation support program 200 and the agent 210 as programs. The storage unit 104 also stores the procedure table 300, the operation instruction index table 400, the group operation instruction index table 500, the operation instruction table 600, the agent table 700, and the LLM service table 800 as data.

[0021] The operational support program 200 is a program that uses agent 210 to support the operational management of the IT system 120.

[0022] Agent 210 is an AI agent that has multiple tools 211 that perform specific processing, and uses the tools 211 and the LLM 151 in the LLM server 150 to autonomously perform tasks based on instructions given to the management computer 100 by a user of the management computer 100 (in this case, the IT system administrator 120). In this embodiment, there are multiple agents 210, but there may be just one. In this embodiment, the instructions given to the management computer 100 are given in the form of instructions written in natural language, similar to instructions given to a human.

[0023] LLM151 is responsible for text interpretation and generation, while tool 211 handles tasks that LLM151 is not good at. Examples of tasks performed by tool 211 include tasks requiring numerical calculations and tasks that retrieve information. Each of the multiple agents 210 may have a different tool 211 and perform different tasks. For example, there may be agents 210 that handle events and agents that handle documents. Furthermore, if there are multiple agents 210, there is a higher-level agent that manages the other agents, and this higher-level agent sequentially operates the other agents in response to instructions from the user to perform the tasks specified in the instructions.

[0024] The procedure table 300 is procedure information that shows procedure manuals describing the procedures for response tasks, which are tasks performed in response to events related to the IT system 120. The procedure manuals are information written in natural language for human use. In this embodiment, the procedure manuals are filed for each response task and stored in the storage unit 104 or an external storage device (not shown).

[0025] The operation instruction index table 400 is index information for searching for agent operation instructions that instruct the operation of agent 210, and for each corresponding task, it shows the agent operation instructions corresponding to that task. Agent operation instructions are an example of support information that assists the operation of agent 210. Note that agent 210 does not necessarily have to operate exactly as instructed by the agent operation instructions, but rather refers to the agent operation instructions to determine its actual operation.

[0026] The group-specific operation instruction index table 500 is another example of index information, and for each work group containing multiple similar corresponding tasks, it shows agent operation instructions for the agent 210 corresponding to that work group.

[0027] The operation instruction table 600 is auxiliary information that shows agent operation instructions corresponding to each corresponding task. The agent table 700 is agent information related to agent 210. The LLM service table 800 is generated AI information related to LLM151.

[0028] Figure 2 shows an example of a procedure table 300. The procedure table 300 shown in Figure 2 includes fields 301 to 304 for each record.

[0029] Field 301 stores path information indicating the location where the file containing the corresponding actions described in the procedure manual is stored, as identification information to identify the corresponding actions described in the procedure manual.

[0030] Fields 302 and 303 store search indexes for finding corresponding tasks. Specifically, field 302 stores a numerical vector as a search index that reflects the meaning of the content of the corresponding task. The numerical vector is generated from the content of the corresponding task, for example, using an embedding process. In this case, the more similar the content of two corresponding tasks, the smaller the distance between their numerical vectors. Field 303 stores keywords included in the content of the corresponding task as a search index.

[0031] Field 304 stores the details of the action to be taken. The action details include the target event, which is the event requiring the action, and the procedure for the action to be taken.

[0032] Figure 3 shows an example of an operation instruction index table 400. The operation instruction index table 400 shown in Figure 3 has fields 401 and 402 for each record.

[0033] Field 401 stores path information, which is identification information for the corresponding task. Field 402 stores an agent operation instruction ID, which identifies the agent operation instruction corresponding to that task.

[0034] Figure 4 shows an example of a group-based operation instruction index table 500. The group-based operation instruction index table 500 shown in Figure 4 has fields 501 and 502 for each record.

[0035] Field 501 stores group path information indicating the location where the files describing the corresponding tasks included in the work group are stored, as identification information to identify a work group containing multiple similar corresponding tasks. In this embodiment, the group path information is represented by a regular expression that encompasses the storage location of the files describing the corresponding tasks included in the work group. Field 502 stores an agent operation instruction ID that identifies the agent operation instruction corresponding to the work group.

[0036] Figure 5 shows an example of an operation instruction table 600. The operation instruction table 600 shown in Figure 5 includes fields 601 to 605 for each record.

[0037] Field 601 stores the agent operation instruction ID, which identifies the agent operation instruction. Fields 602 to 605 store the instruction details of the agent operation instruction identified by the agent operation instruction ID in the same record. Specifically, field 602 stores a number indicating the execution order of agent 210. Field 603 stores the agent name, which is the name of the agent, as identification information to identify the agent 210 that is instructed to be executed by each number. Field 604 stores the tool name, which is the name of the tool, as identification information to identify the tool 211 that is instructed to be used by the agent that is instructed to be executed. Field 605 stores skipability information, which indicates whether the execution of the tool that is instructed to be used can be skipped. The skipability information indicates "Yes" if skipping is possible and "No" if skipping is not possible.

[0038] Figure 6 shows an example of agent table 700. The agent table 700 shown in Figure 6 contains fields 701 to 703 for each record.

[0039] Field 701 stores the agent name of agent 210. Field 702 stores the agent prompt, which is a descriptive text describing the contents of agent 210. The agent prompt is information that allows LLM151 to understand the contents of agent 210. Field 703 stores tool information about tool 211 that agent 210 possesses.

[0040] Field 703 specifically includes fields 704 to 706. Field 704 stores the tool name of tool 211. Field 705 stores the tool prompt, which is a descriptive text explaining the contents of tool 211. The tool prompt is information that allows LLM151 to understand the contents of tool 211. Field 706 stores the function that is the actual tool.

[0041] Figure 7 is a diagram illustrating an example of an LLM service table 800. The LLM service table 800 shown in Figure 7 contains fields 801 to 803 for each record.

[0042] Field 801 stores the LLM name, which is the name of LLM151, as identification information for LLM151. Field 802 stores the service URL (Uniform Resource Locator) indicating the network location of LLM151. Field 803 stores the API (Application Programming Interface) key (KEY) for using LLM.

[0043] Figure 8 is a diagram illustrating an example of task execution processing by the management computer 100. Task execution processing is the process by which the AI ​​agent 201 executes a task based on instructions from the user. In this process, when the operation support program 200 references the work content in the procedure table 300 during task processing, it provides the AI ​​agent 201 with agent operation instructions corresponding to the target work, which is the corresponding work to be performed according to that work content, thereby causing the AI ​​agent 201 to execute the task based on the agent operation instructions.

[0044] Specifically, the IT system administrator 121, who is a user of the management computer 100, first inputs an instruction message to the management computer 100 (step S1).

[0045] Next, the operation support program 200 of the management computer 100 determines whether or not the work content in the procedure table 300 is referenced in the task based on the instruction (step S2). Examples of processes that reference work content include the process of searching for work content in the procedure table 300.

[0046] If the task details are not referenced (Step S2: No), the operation support program 200 has agent 210 execute the task (Step S3) and then terminates the process.

[0047] On the other hand, if the work content is referenced (Step S2: Yes), the operation support program 200 determines whether or not there is an agent operation instruction corresponding to the referenced work, which is the corresponding work referenced in the task (Step S4).

[0048] If an agent operation instruction exists (Step S4: Yes), the operation support program 200 operates agent 210 based on the instruction statement, reference operation, and agent operation instruction (Step S5).

[0049] If no agent operation instruction exists (Step S4: No), the operation support program 200 creates an agent operation instruction corresponding to the reference operation based on the similar operation instruction, which is an agent operation instruction corresponding to a similar operation that is similar to the reference operation (Step S6). Then, the operation support program 200 operates agent 210 based on the created agent operation instruction, instruction statement and reference operation (Step S7).

[0050] Agent 210, based on the task, reference work, and agent operation instructions, uses tool 211 and LLM 151 to execute processing according to the task, and outputs the execution results to the operation manager 121, for example, using display 105.

[0051] The task execution process described above is merely an example and is not limited to it. For example, if no agent operation instruction exists (step S4: No), the operation support program 200 may operate agent 210 based on the instruction statement and reference operations without creating a new agent operation instruction.

[0052] Figure 9 is a flowchart illustrating a more detailed example of the task execution process. Here, we explain an example where no new agent action instructions are created when no agent action instructions exist.

[0053] First, the operation support program 200 of the management computer 100 receives event information from the operations manager 121 indicating an event related to the IT system 120, along with instructions related to that event (step S100). The instructions are natural language, such as, "Read the procedure manual corresponding to the selected event, determine the severity of the event, and if it is severe, contact a higher-level operations manager."

[0054] The operation support program 200 determines whether the task based on the received instruction has been completed (step S101). If the task processing is completed (step S101: Yes), the operation support program 200 terminates its operation.

[0055] If the task is not completed (Step S101: No), the higher-level agent among Agent 210 that manages the other agents determines, based on the instructions and event information, which agent will perform the next processing in the task (Step S102).

[0056] Based on the task and event information, the target agent determines which tool to use from among its own tools 211, executes that target tool, and performs the processing related to the task (step S103).

[0057] The operational support program 200 determines whether the target tool is a procedure manual search tool that searches for corresponding tasks within the procedure manual (step S104). If the target tool is not a procedure manual search tool (step S104: No), the operational support program 200 returns to the process in step S101.

[0058] If the target tool is a procedure manual search tool (Step S104: Yes), the operation support program 200 obtains path information for the corresponding task found by the procedure manual search tool (Step S105). For example, if the procedure manual search tool is a tool that uses RAG (Retrieval-Augmented Generation) to search for corresponding tasks using a generated AI model (e.g., LLM151) to search for documents, when a corresponding task is found, the path information of the search source from which that corresponding task was found is obtained. Therefore, the operation support program 200 can obtain path information from the searched tool.

[0059] The operation support program 200 determines in the operation instruction index table 400 whether or not a corresponding ID, which is an agent operation instruction ID corresponding to the acquired path information, exists (step S106).

[0060] If no corresponding ID exists (Step S106: No), the operation support program 200 determines in the group-specific operation instruction index table 500 whether a match ID exists, which is an agent operation instruction ID corresponding to group path information that matches the acquired path information (Step S107). Specifically, the group path information that matches the path information is group path information described using a regular expression that includes the path information.

[0061] If no match ID exists (step S107: No), the operation support program 200 returns to the process in step S101.

[0062] If a corresponding ID exists (step S106: Yes), or if a match ID exists (step S107: Yes), the operation support program 200 retrieves an agent operation instruction corresponding to that corresponding ID or match ID from the operation instruction table 600 (step S108). The operation support program 200 adds the retrieved agent operation instruction to the instruction statement (step S109) and returns to the process in step S101. This allows the agent 210 to execute the rest of the subsequent tasks based on the instruction statement containing the agent operation instruction.

[0063] Figure 10 is a flowchart illustrating another example of a more detailed task execution process. Here, we illustrate an example where a new agent action instruction is created when no agent action instruction exists.

[0064] The task execution process in Figure 10 differs from the task execution process described using Figure 9 in that, in step S106, if no corresponding ID exists (step S106: No), the operation support program 200 executes a creation process (step S200) to create a new agent operation instruction corresponding to the corresponding task, instead of the process in step S107. Note that if the process in step S200 is completed, the process in step S108 is executed.

[0065] Figure 11 is a flowchart illustrating an example of the creation process in step S200 of Figure 10.

[0066] In the creation process, first, the operation support program 200 obtains path information for the first task, which is the corresponding task for which the agent operation instruction is to be created (step S201). For example, the operation support program 200 obtains the path information obtained in step S105 of Figure 10 as the path information for the first task.

[0067] The operational support program 200 retrieves the work content corresponding to the path information of the first task from the procedure table 300 as the work content of the first task (step S202).

[0068] The operational support program 200 identifies a second task, which is a corresponding task having similar work content to the first task, based on the procedure table 300 (step S203).

[0069] Specifically, the operational support program 200 compares the search index of the first task with the search index of other tasks, calculates the similarity between the search index of each other task and the search index of the first task, and identifies the second task based on that similarity. For example, the operational support program 200 may identify a predetermined number of other tasks as the second task, starting with those with the highest similarity, or it may identify other tasks as the second task if their similarity is above a predetermined threshold. The similarity is, for example, the distance between numerical vectors if the search index is a numerical vector, or the degree of keyword matching if the search index is a keyword. Alternatively, the similarity may be a value based on the distance between numerical vectors and the degree of keyword matching (for example, a weighted sum).

[0070] Once the second task is identified, the operational support program 200 determines whether there are any unselected second tasks that have not yet been chosen (step S204).

[0071] If there is a second unselected task (step S204: Yes), the operation support program 200 selects one of the second unselected tasks as the selected task (step S205). The operation support program 200 then determines in the operation instruction index table 400 whether or not a second agent operation instruction ID exists that corresponds to the selected task (step S206).

[0072] If a second agent operation instruction ID does not exist (step S206: No), the operation support program 200 returns to the process in step S204.

[0073] On the other hand, if a second agent operation instruction ID exists (step S206: Yes), the operation support program 200 retrieves the agent operation instruction corresponding to the second agent operation instruction ID from the operation instruction table 600 as a selection operation instruction, which is an agent operation instruction corresponding to the selection operation (step S207).

[0074] The operation support program 200 generates hint information that associates the selection task with the selection operation instruction (step S208), and then returns to the process in step S204.

[0075] Furthermore, if there is no unselected second task (step S204: No), the operation support program 200 creates an agent operation instruction corresponding to the first task based on the hint information (step S209), and then terminates the creation process. For example, the operation support program 200 inputs hint information and a prompt to the LLM151 indicating that an agent operation instruction should be created based on the hint information, thereby causing the LLM151 to create an agent operation instruction corresponding to the first task.

[0076] When the creation process is complete, the operation support program 200 executes the process in step S108 of Figure 10. At that time, instead of obtaining the agent operation instruction corresponding to the corresponding ID or match ID from the operation instruction table 600, the operation support program 200 obtains the agent operation instruction created in step S209 of Figure 11.

[0077] Figure 12 is a flowchart illustrating an example where the creation process is performed at a different time. In the examples in Figures 10 and 11, agent operation instructions were created when the instruction text was entered, but in the example in Figure 12, agent operation instructions are created in advance for all corresponding tasks.

[0078] First, the operational support program 200 acquires all the corresponding tasks shown in the procedure table 300 at a predetermined timing (step S301). The predetermined timing is, for example, the timing when the user instructs the creation of an agent operation instruction.

[0079] The operational support program 200 determines whether there are any unselected tasks among the tasks that have not yet been selected (step S302).

[0080] If there are unselected corresponding tasks (Step S302: Yes), the operation support program 200 selects one of the unselected tasks as a selected task (Step S303). The operation support program 200 then determines in the operation instruction index table 400 whether or not there is a corresponding operation instruction, which is an agent operation instruction, that corresponds to the selected task (Step S304).

[0081] If a corresponding operation instruction exists (step S304: No), the operation support program 200 returns to the process in step S302.

[0082] On the other hand, if no corresponding operation instruction exists (step S304: Yes), the operation support program 200 executes a creation process to create an agent operation instruction corresponding to the selected task (step S200).

[0083] When an agent operation instruction is created during the creation process, the operation support program 200 adds the selected task and the created agent operation instruction to the operation instruction table 600, and also adds the path information of the selected task and the agent operation instruction ID of the created agent operation instruction to the operation instruction index table 400 (step S305), and then returns to the process in step S302.

[0084] As described above, according to this embodiment, the storage unit 104 stores a procedure table 300 which describes the procedures for corresponding tasks for each task in response to an event, and an operation instruction table 600 which shows agent operation instructions for each corresponding task, which are support information that assists the operation of agent 210. When the CPU 101 references a work procedure in the procedure table 300 in a task based on an instruction statement, it provides agent 210 with an agent operation instruction corresponding to the target task, which is a corresponding task to be performed according to that work procedure, and causes agent 210 to execute the task based on the agent operation instruction. Therefore, it is possible to make agent 210 perform processing in accordance with the procedures for corresponding tasks, and thus it is possible to make the tasks in response to events perform appropriately.

[0085] Furthermore, in this embodiment, if the CPU 101 references a work procedure in the procedure table 300 during a task, it provides an agent operation instruction to the agent 210. This makes it possible to execute processing in accordance with the procedure of the work corresponding to the agent operation instruction as needed.

[0086] Furthermore, in this embodiment, the agent operation instruction indicates the execution order of multiple agents 210. In this case, it becomes possible to execute multiple agents 210 in an appropriate order.

[0087] Furthermore, in this embodiment, the agent operation instruction indicates the tool 211 that agent 210 will use in the task. Agent 210 can be made to execute the appropriate tool.

[0088] Furthermore, in this embodiment, for a first task, which is a corresponding task for which no agent operation instruction exists, the CPU 101 identifies a second task, which is a corresponding task similar to the first task, based on the procedure table 300, and generates support information corresponding to the first task based on the agent operation instruction corresponding to the second task. In this case, even for corresponding tasks for which no agent operation instruction has been prepared in advance, the agent 210 can be made to execute the task appropriately.

[0089] Furthermore, in this embodiment, the CPU 101 searches for a corresponding task for which no manual agent operation instruction exists at a predetermined timing and generates an agent operation instruction corresponding to that task. Therefore, since it is possible to generate an agent operation instruction before actually executing the task, it is possible to reduce the processing load during the task.

[0090] The embodiments of the Disclosure described above are illustrative for illustrative purposes and are not intended to limit the scope of the Disclosure to those embodiments only. Those skilled in the art can implement the Disclosure in various other forms without departing from the scope of the Disclosure. [Explanation of Symbols]

[0091] 100: Management computer 101: CPU 102: Memory 103: Network interface 104: Storage unit 105: Display 120: IT system 121: Operations administrator 130: Cluster 131: Node 140: Storage 150: LLM server 200: Operation support program 210: Agent 211: Tool

Claims

1. A work support device that assists the aforementioned work using an AI agent that performs tasks based on instructions related to work in response to an event, A storage unit that stores procedure information describing the steps of the aforementioned work for each of the aforementioned work, and auxiliary information indicating support information for assisting the operation of the AI ​​agent for each of the aforementioned work, A work support device comprising: a control unit that, when the procedure in the procedure manual information is referenced in the aforementioned task, provides the AI ​​agent with support information corresponding to the target task which is the work performed in accordance with said procedure, and causes the AI ​​agent to execute the task based on the support information.

2. The work support device according to claim 1, wherein the control unit, when the procedure in the procedure information is referenced in the middle of the task, provides the support information to the AI ​​agent and causes the AI ​​agent to continue the task based on the support information.

3. There are multiple AI agents, The work support device according to claim 1, wherein the support information indicates the execution order of the plurality of AI agents.

4. The AI ​​agent has one or more tools that perform specific processing, The work support device according to claim 1, wherein the support information indicates the tools used by the AI ​​agent in the task.

5. The work support device according to claim 1, wherein the control unit identifies a second work that is similar to the first work based on the procedure manual information for a first work for which no support information exists in the auxiliary information, and generates support information corresponding to the first work based on the support information corresponding to the second work.

6. The work support device according to claim 5, wherein the control unit generates the support information for the target work as the first work if no support information corresponding to the target work exists.

7. The work support device according to claim 5, wherein the control unit searches for the first work from the procedure information at a predetermined timing and generates support information corresponding to the first work.

8. The work support device according to claim 5, wherein the control unit generates the support information using a generated AI model.

Images