Agent-based enterprise task execution method and device, equipment and storage medium

Abstract

This application discloses an enterprise task execution method, apparatus, device, and storage medium based on intelligent agents, relating to the field of intelligent agent technology. The method includes: determining a structured target planning domain based on use cases in user requests and a preset planning domain library; obtaining real-time business data from the enterprise backend system through a business system interface; using intelligent agents to parse, organize, and semantically transform the real-time business data to obtain data structure information and semantic representation information conforming to the target planning domain definition; mapping the real-time business data to the target planning domain to obtain the current system business state, thereby constructing a structured reasoning context, and sending it to a large language model service to obtain a task planning sequence; performing behavior pruning on the task planning sequence to obtain a set of safe and executable actions; and calling the corresponding virtual capability set through the capability virtualization layer of a preset capability governance center to execute each task and obtain the enterprise task execution result, thereby improving the efficiency of enterprise task execution using intelligent agents.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent technology, and in particular to an enterprise task execution method, apparatus, device and storage medium based on intelligent agents. Background Technology

[0002] Currently, in enterprise-level application scenarios, intelligent agent systems need to possess the capabilities of understanding user requests, reasoning about business semantics, planning tasks, and executing operations under control to support the automated processing needs of complex business systems. Existing agent systems typically complete task processing based on prompt word engineering, tool call interfaces, and retrieval enhancement generation mechanisms. The model uses natural language prompts for intent recognition and task planning, and directly maps the inference results to system operation instructions. In the above technical solutions, intent recognition, task planning, and execution processes are highly coupled, and the inference results directly drive subsequent operations. The system level lacks constraints on the inference path and execution decisions, and the boundaries of agent behavior are implicit in the model inference process, making runtime control and behavior auditing difficult.

[0003] Enterprise business rules, process logic, and compliance requirements are often embedded in natural language using prompts or context in model reasoning. This lack of structured expression makes unified management, version control, and change implementation difficult. The scope and constraints of business knowledge are unclear, failing to meet enterprise-level governance requirements. For scenarios involving high-risk or irreversible business operations, existing agent systems lack robust risk identification and execution restriction mechanisms. Deviations in reasoning results may directly translate into business actions, and the system lacks effective security protection measures.

[0004] Existing agent systems are mostly built around model reasoning and tool invocation, lacking capability governance and interface constraint design for enterprise-level applications, making it difficult to form stable collaboration with enterprise permission systems, business process systems and back-end business systems.

[0005] With the application of Large Language Models (LLMs) and Agent technology in enterprise scenarios, while existing technologies can achieve task planning and tool invocation, fundamental defects still exist at the runtime governance level, mainly reflected in: 1. Uncontrollable Agent Behavior: Existing agent systems typically enter the execution process after completing task decomposition during the inference phase. They lack the ability to dynamically negate, trim, and backtrack on decisions made during operation. Once a decision is generated, it is difficult to intercept it at the system level before or during execution, resulting in high-risk operations being uncontrollable.

[0006] 2. Business knowledge injection is unmanageable: In existing solutions, business rules, processes and constraints are mostly injected into the model in the form of natural language prompts or search results. This makes it impossible to form runtime constraint entities independent of the model's inference results. As a result, business rules are difficult to version manage, their effective path cannot be audited, and the behavior of agents cannot be forcibly constrained during operation.

[0007] 3. Lack of reverse governance capabilities in the "decision-execution-audit" closed loop: Existing technologies usually treat auditing as a means of post-event recording. Audit results cannot influence subsequent decision generation, resulting in the system's inability to achieve dynamic governance and strategy evolution based on historical risk behaviors.

[0008] As can be seen from the above, how to improve the efficiency of using intelligent agents to perform enterprise tasks is an urgent problem to be solved. Summary of the Invention

[0009] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for executing enterprise tasks based on intelligent agents, which can improve the efficiency of enterprise task execution using intelligent agents. The specific solution is as follows: Firstly, this application provides an agent-based enterprise task execution method, including: Based on user use cases in user requests and a preset planning domain library, a structured target planning domain is determined, and then real-time business data corresponding to the enterprise back-end system is obtained through the business system interface; the target planning domain is an AI-reasonable planning domain. Based on the requirements of the target planning domain, real-time business data corresponding to the target planning domain is obtained through the business system interface. The intelligent agent is used to parse, normalize and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, so as to determine the real-time business data based on the data structure information and semantic representation information. The real-time business data is mapped to the target planning domain using an intelligent agent to obtain the current system business state. A structured inference context is then constructed based on the target planning domain and the current system business state. The inference context is then sent to a preset large language model service to obtain a task planning sequence. Based on the target planning domain, the task planning sequence is pruned to obtain a set of safe and executable actions. Then, the agent is used to call the set of virtual capabilities corresponding to the target planning domain based on the capability virtualization layer in the preset capability governance center, so as to execute each task in the set of actions based on the set of virtual capabilities and obtain the corresponding enterprise task execution results.

[0010] Optionally, determining the structured target planning domain based on user use cases in user requests and a preset planning domain library includes: The corresponding user use case identifier is parsed from the user request, and a multi-dimensional matching search is performed on the user use case identifier in the preset planning domain library to obtain the matching search results; the preset planning domain library is a library stored using a graph database structure; the preset planning domain library includes nodes corresponding to each planning domain; the node attributes include business objectives, business state sets, action models, and constraint rules; the nodes are connected by edges; The semantic similarity between the user use case identifier and each node is determined by using a graph traversal algorithm and based on user role attributes and real-time system context. The nodes with the highest similarity among the semantic similarities are selected as a candidate set using a preset historical execution success rate. By using a preset dynamic adjustment module and based on real-time business environment data and the matching retrieval results, the threshold of state variables and priority of constraint rules corresponding to the candidate set are adaptively calibrated to obtain a structured target planning domain.

[0011] Optionally, the step of obtaining real-time business data corresponding to the target planning domain through the business system interface based on the requirements of the target planning domain, and using the intelligent agent to parse, normalize, and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, and determining real-time business data based on the data structure information and semantic representation information, includes: Determine the requirements corresponding to the target planning domain, and build a business system interface with a distributed microservice architecture, data cleaning and normalization pipeline. Then, use an asynchronous message queue to parse the first business data to be processed, which includes several data change events. Using the business system interface and based on preset data quality rules and the requirements, the first business data to be processed is sequentially subjected to noise reduction, deduplication and format verification operations to obtain heterogeneous business data. The heterogeneous business data is regularized to obtain structured second business data to be processed. The second business data to be processed is then subjected to window aggregation, state calculation and semantic transformation using a real-time stream processing engine to obtain a real-time business data snapshot. The real-time business data snapshot includes the original data values ​​and business indicators. The real-time business data snapshot is sent to the intelligent agent through a preset interface using the business system interface, so that the intelligent agent can determine the real-time business data based on the real-time business data snapshot.

[0012] Optionally, the step of using an intelligent agent to map the real-time business data to the target planning domain to obtain the current system business state, and constructing a structured reasoning context based on the target planning domain and the current system business state, includes: The mapping relationship matrix is ​​determined by utilizing the integrated state mapping engine and deep learning model in the intelligent agent, and based on the preset knowledge graph, the real-time business data and the state variables of the target planning domain. The current system business state corresponding to each state variable is determined by using a preset state mapping engine and based on the mapping relationship matrix. Then, the business objectives, action model subsets and constraint rules corresponding to the target planning domain are extracted by using a preset inference context construction module and based on the mapping relationship matrix. A structured reasoning context is generated based on the business objective, the action model subset and the constraint rules, similar case features in the preset historical execution library, the target planning domain and the current system business state; the reasoning context includes time sequence markers and priority labels.

[0013] Optionally, sending the inference context to a preset large language model service to obtain a task planning sequence includes: The intelligent agent is used to send the inference context to a preset large language model service via an API gateway; the API gateway is used for data flow control and anomaly detection. The inference context is processed using the domain-specific language model in the preset large language model service and based on the preset prompt word project to obtain the initial task planning sequence; The initial task planning sequence is merged, parallelized, and resource-dependently analyzed using a pre-defined task planning sequence post-processing center and a pre-defined graph optimization algorithm to obtain the target task planning sequence. The target task planning sequence is in the form of a directed acyclic graph. The nodes in the target task planning sequence are atomic actions, and the edges are state transition conditions.

[0014] Optionally, after performing behavior pruning on the task planning sequence based on the target planning domain to obtain a safe and executable action set, the method further includes: A set of atomic action models, a set of state transition rules, and a set of constraint rules are loaded from the target planning domain using a rule engine, so as to construct a pruning rule library based on the set of atomic action models, the set of state transition rules, and the set of constraint rules; Each task action in the task planning sequence is converted into a temporal logic formula, and the pruning rule base is used to verify whether the task planning sequence satisfies the safety and liveness attributes. If the verification passes, the planning verifier is used to determine whether each task action in the task planning sequence belongs to the set of atomic action models corresponding to the target planning domain. If it does, it is determined whether the state before and after the execution of each task action follows the state transition rules of the target planning domain. If it does, it is determined whether the execution state corresponding to each task action satisfies the set of constraint rules of the target planning domain. If satisfied, each task planning sequence is set as a set of secure and executable actions, and a signature certificate corresponding to the task planning sequence is generated using the planning verifier. If not satisfied, a dynamic correction sub-process is triggered, and correction suggestions are generated based on the rule engine. Then, the suggestions and error codes are returned to the agent through the feedback interface to drive the iterative optimization of the task planning sequence.

[0015] Optionally, the step of pruning the task planning sequence based on the target planning domain to obtain a set of safe and executable actions, and then using the intelligent agent and based on the capability virtualization layer in the preset capability governance center to call the set of virtual capabilities corresponding to the target planning domain, to execute each task in the set of actions based on the set of virtual capabilities, and to obtain the corresponding enterprise task execution result, includes: The intelligent agent sends a capability invocation request to a preset capability governance center; the preset capability governance center includes a capability virtualization layer; the capability virtualization layer is used to uniformly encapsulate and abstract enterprise system interfaces, and maintain a registry including atomic action models and system tools bound to each planning domain; the capability invocation request includes an action identifier, a parameter list, and a session token; the intelligent agent only supports accessing the virtual capability set corresponding to the current planning domain during runtime; the virtual capability set supports unified scheduling and auditing records using the capability virtualization layer; The system utilizes a preset authentication center and verifies whether the user role has the authority to act, whether there are quota restrictions, and the availability of system tool resources based on the capability call request. After the verification is passed, the preset capability governance center is invoked to call the corresponding system tools or interfaces to monitor the execution process of the task and collect real-time indicators to obtain the enterprise task execution results. After the task is completed, the preset capability governance center is invoked to return the enterprise task execution result to the intelligent agent, so that the intelligent agent can update the current system business status based on the enterprise task execution result and record the execution trace to the audit log; An audit report is generated by using a pre-defined unsupervised machine learning algorithm and analyzing the execution patterns based on the audit logs, identifying deviations from expected behavior and performance bottlenecks; the audit report includes a compliance score, an efficiency heatmap, and root cause analysis.

[0016] Secondly, this application provides an enterprise task execution device based on an intelligent agent, comprising: The target planning domain determination module is used to determine a structured target planning domain based on user use cases in user requests and a preset planning domain library, and then obtain real-time business data corresponding to the enterprise back-end system through the business system interface; the target planning domain is an AI-reasonable planning domain. The business data determination module is used to obtain real-time business data corresponding to the target planning domain through the business system interface based on the requirements of the target planning domain, and use the intelligent agent to parse, normalize and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, so as to determine the real-time business data based on the data structure information and semantic representation information. The inference context determination module is used to map the real-time business data to the target planning domain using an intelligent agent to obtain the current system business state, and to construct a structured inference context based on the target planning domain and the current system business state. Then, the inference context is sent to a preset large language model service to obtain a task planning sequence. The enterprise task execution result determination module is used to perform behavior trimming on the task planning sequence based on the target planning domain to obtain a safe and executable action set. Then, it uses the intelligent agent and the capability virtualization layer in the preset capability governance center to call the virtual capability set corresponding to the target planning domain, so as to execute each task in the action set based on the virtual capability set and obtain the corresponding enterprise task execution result.

[0017] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned agent-based enterprise task execution method.

[0018] Fourthly, this application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned agent-based enterprise task execution method.

[0019] As can be seen from the above, before executing enterprise tasks based on intelligent agents, this application needs to determine a structured target planning domain based on user use cases in user requests and a preset planning domain library. Then, it obtains real-time business data corresponding to the enterprise backend system through the business system interface. The target planning domain is a planning domain that AI can reason about. Based on the requirements of the target planning domain, it obtains real-time business data corresponding to the target planning domain through the business system interface, and uses intelligent agents to parse, organize, and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain. Based on the data structure information and semantic representation information, it determines the real-time business data. It uses intelligent agents to map the real-time business data to the target planning domain to obtain the current system business state, and constructs a structured reasoning context based on the target planning domain and the current system business state. Then, it sends the reasoning context to the preset large language model service to obtain the task planning sequence. Based on the target planning domain, it performs behavior trimming on the task planning sequence to obtain a safe and executable action set. Then, it uses intelligent agents and the capability virtualization layer in the preset capability governance center to call the virtual capability set corresponding to the target planning domain, and executes each task in the action set based on the virtual capability set to obtain the corresponding enterprise task execution result.

[0020] Therefore, this application first needs to determine a structured target planning domain based on user use cases in user requests and a preset planning domain library. Then, it obtains real-time business data corresponding to the enterprise backend system through the business system interface. The target planning domain is an AI-reasonable planning domain. Second, based on the requirements of the target planning domain, it obtains real-time business data corresponding to the target planning domain through the business system interface, and uses an intelligent agent to parse, organize, and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain. Based on the data structure information and semantic representation information, it determines the real-time business data. Then, it uses an intelligent agent to map the real-time business data to the target planning domain to obtain the current system business state. Based on the target planning domain and the current system business state, it constructs a structured reasoning context and sends the reasoning context to a preset large language model service to obtain a task planning sequence. Finally, it performs behavior trimming on the task planning sequence based on the target planning domain to obtain a safe and executable action set. Then, it uses an intelligent agent and the capability virtualization layer in the preset capability governance center to call the virtual capability set corresponding to the target planning domain to execute each task in the action set based on the virtual capability set, and obtains the corresponding enterprise task execution result. This improves the efficiency of using intelligent agents to perform enterprise tasks, thereby enhancing the user experience. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This application discloses a flowchart of an enterprise task execution method based on intelligent agents. Figure 2 This is a schematic diagram of the structure of an enterprise task execution device based on an intelligent agent disclosed in this application; Figure 3 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Currently, in enterprise-level application scenarios, intelligent agent systems need to possess the capabilities of understanding user requests, reasoning about business semantics, planning tasks, and executing controlled operations to support the automated processing needs of complex business systems. Enterprise business rules, process logic, and compliance requirements are often embedded in natural language using prompts or context in model reasoning, lacking structured expression methods. This makes unified management, version control, and change effectiveness difficult, and the scope and constraints of business knowledge are unclear, failing to meet enterprise-level governance requirements. For scenarios involving high-risk or irreversible business operations, existing agent systems lack robust risk identification and execution restriction mechanisms; biases in reasoning results may directly translate into business operational behaviors, and the system lacks effective security protection measures. Therefore, this application provides an agent-based enterprise task execution method that improves the efficiency of using agents for enterprise task execution.

[0025] See Figure 1 As shown, this embodiment of the invention discloses an enterprise task execution method based on an intelligent agent, including: Step S11: Determine a structured target planning domain based on user use cases in user requests and a preset planning domain library, and then obtain real-time business data corresponding to the enterprise back-end system through the business system interface; the target planning domain is an AI-reasonable planning domain.

[0026] In this embodiment, the present application aims to address the common problems of uncontrollable behavior and unmanageable business knowledge injection in existing agents in enterprise scenarios. It is worth noting that, unlike existing solutions centered on Prompt or process orchestration, this embodiment, from an enterprise-level application perspective, confines the agent's behavioral capabilities to a reasonable, constrained, and manageable business planning domain. This ensures that the agent's behavior is no longer directly determined by model inference results, but is constrained by the business structure and capability boundaries defined by the system.

[0027] It is worth mentioning that the core design idea of ​​this application embodiment is: to take the user case as the smallest business unit; to model each user case as an AI reasoning planning domain; to uniformly govern the agent's business knowledge understanding and executable capability range through the planning domain; and to ensure that the agent's decision-making, reasoning and execution processes always run within a clear business boundary. In this way, the agent's controllable execution and stable deployment in enterprise-level scenarios are achieved.

[0028] Furthermore, to fundamentally address the issue of unmanageable business knowledge injection, this application proposes an AI-based reasoning planning domain modeling method based on UserCase. The planning domain describes the complete business world model of the Agent under a specific business use case. It is not a prompt template, but rather a structured business constraint space that the Agent can use for reasoning and decision-making. Notably, each planning domain in this application includes at least the following core elements: Goal State: Defines the final business state expected to be achieved by the user use case; Business State: Describes the legal states and their transition relationships during business execution using a state machine approach, constraining the Agent's behavioral sequence and execution path; Action Model: Defines the permitted business operation capabilities within the planning domain and binds them to specific system tools or interfaces; Constraint Rules: Describes permission boundaries, compliance requirements, risk control rules, and preconditions, serving as mandatory constraints for Agent reasoning and execution.

[0029] In this way, through the above modeling method, business knowledge is no longer injected into the model in the form of natural language prompts, but participates in the agent's reasoning and behavior control in the form of a structured planning domain, thereby realizing the manageability, reusability and evolvability of business knowledge.

[0030] Specifically, determining the structured target planning domain based on user use cases in user requests and a pre-defined planning domain library can include: parsing the corresponding user use case identifier from the user request, and performing multi-dimensional matching and retrieval based on the user use case identifier in the pre-defined planning domain library to obtain matching and retrieval results; the pre-defined planning domain library is a library stored using a graph database structure; the pre-defined planning domain library includes nodes corresponding to each planning domain; node attributes include business objectives, business state sets, action models, and constraint rules; each node is connected by edges; using a graph traversal algorithm and based on user role attributes and real-time system context to determine the semantic similarity between user use case identifiers and each node, and using a pre-defined historical execution success rate to select the nodes with the highest semantic similarity as a candidate set; using a pre-defined dynamic adjustment module and based on real-time business environment data and matching and retrieval results to adaptively calibrate the threshold of state variables and the priority of constraint rules corresponding to the candidate set to obtain the structured target planning domain.

[0031] Step S12: Based on the requirements of the target planning domain, obtain real-time business data corresponding to the target planning domain through the business system interface, and use the intelligent agent to parse, normalize and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, so as to determine the real-time business data based on the data structure information and semantic representation information.

[0032] In this embodiment, based on the requirements of the target planning domain, real-time business data corresponding to the target planning domain is obtained through the business system interface. The intelligent agent then parses, normalizes, and semantically transforms the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain. Determining real-time business data based on the data structure information and semantic representation information may include: determining the requirements corresponding to the target planning domain, constructing a business system interface with a distributed microservice architecture and data cleaning and normalization pipeline, and then using an asynchronous message queue to parse the first pending business data including several data change events; utilizing the business system... The system interface, based on preset data quality rules and the requirements, sequentially performs noise reduction, deduplication, and format verification operations on the first business data to be processed to obtain heterogeneous business data; the heterogeneous business data is then regularized to obtain structured second business data to be processed, and a real-time stream processing engine is used to perform window aggregation, state calculation, and semantic transformation on the second business data to obtain a real-time business data snapshot; the real-time business data snapshot includes the original data values ​​and business indicators; the real-time business data snapshot is then sent to the intelligent agent through a preset interface using the business system interface, so that the intelligent agent can determine the real-time business data based on the real-time business data snapshot.

[0033] Step S13: Use an intelligent agent to map the real-time business data to the target planning domain to obtain the current system business state, and construct a structured inference context based on the target planning domain and the current system business state. Then, send the inference context to the preset large language model service to obtain the task planning sequence.

[0034] In this embodiment, to address the issue of uncontrollable Agent behavior, this application introduces a behavior governance mechanism based on planning domains during system runtime to ensure that all Agent behaviors are constrained by the currently effective planning domain. This mechanism includes at least the following steps: determining the set of currently available planning domains based on user requests, user roles, and system context; the Agent performing task reasoning and behavior planning only within the selected planning domains. Furthermore, the candidate behaviors generated by the Agent must meet the following requirements: belonging to the action model defined within the planning domain; conforming to the current business state and state transition rules; not violating the constraint rules in the planning domain; rejecting or substituting behaviors that do not meet the planning domain constraints; and only allowing behaviors that pass the planning domain validation to enter the execution phase. Through this mechanism, this application transforms Agent behavior from "model-driven" to "planning domain-constrained," ensuring the determinism and controllability of Agent behavior at the system level.

[0035] Specifically, the process involves using an intelligent agent to map real-time business data to a target planning domain to obtain the current system business state. A structured inference context is then constructed based on the target planning domain and the current system business state. This can include: utilizing an integrated state mapping engine and deep learning model within the intelligent agent, and determining a mapping matrix based on a pre-defined knowledge graph, real-time business data, and state variables of the target planning domain; using the pre-defined state mapping engine and the mapping matrix to determine the current system business state corresponding to each state variable; then using a pre-defined inference context construction module and the mapping matrix to extract business objectives, action model subsets, and constraint rules corresponding to the target planning domain; and generating a structured inference context based on the business objectives, action model subsets, constraint rules, similar case features from a pre-defined historical execution library, the target planning domain, and the current system business state. The inference context includes time-series markers and priority labels.

[0036] Furthermore, sending the inference context to a pre-defined large language model service to obtain a task planning sequence can include: using an agent and an API (Application Programming Interface) gateway to send the inference context to the pre-defined large language model service; the API gateway is used for data flow control and anomaly detection; using the domain-specific language model in the pre-defined large language model service and based on a pre-defined prompt word engineering to process the inference context to obtain an initial task planning sequence; using a pre-defined task planning sequence post-processing center and based on a pre-defined graph optimization algorithm to merge, parallelize scheduling, and perform resource dependency analysis on the initial task planning sequence to obtain a target task planning sequence; the target task planning sequence is in the form of a directed acyclic graph; the nodes in the target task planning sequence are atomic actions, and the edges are state transition conditions.

[0037] Step S14: Use an intelligent agent to map the real-time business data to the target planning domain to obtain the current system business state, and construct a structured inference context based on the target planning domain and the current system business state. Then, send the inference context to the preset large language model service to obtain the task planning sequence.

[0038] In this embodiment, the application requires separate determination of each task action in the task planning sequence. Specifically, after sending the inference context to a preset large language model service to obtain the task planning sequence, the method further includes: using a rule engine to load an atomic action model set, a state transition rule set, and a constraint rule set from the target planning domain, and constructing a pruning rule base based on the atomic action model set, state transition rule set, and constraint rule set; converting each task action in the task planning sequence into a temporal logic formula, and using the pruning rule base to verify whether the task planning sequence satisfies the safety and liveness attributes; if the verification passes, using a planning verifier to determine the task. If each task action in the planning sequence belongs to the set of atomic action models corresponding to the target planning domain, then it is determined whether the state before and after the execution of each task action follows the state transition rules of the target planning domain. If it does, then it is determined whether the execution state corresponding to each task action satisfies the set of constraint rules of the target planning domain. If it does, then each task planning sequence is set as a set of safe and executable actions, and a signature certificate corresponding to the task planning sequence is generated using the planning verifier. If it does not satisfy, then a dynamic correction subprocess is triggered, and correction suggestions are generated based on the rule engine. Then, the suggestions and error codes are returned to the agent through the feedback interface to drive the iterative optimization of the task planning sequence.

[0039] Step S15: Based on the target planning domain, perform behavior trimming on the task planning sequence to obtain a safe and executable action set. Then, use the intelligent agent and the capability virtualization layer in the preset capability governance center to call the virtual capability set corresponding to the target planning domain, so as to execute each task in the action set based on the virtual capability set and obtain the corresponding enterprise task execution result.

[0040] In this embodiment, the present application requires the use of intelligent agents and a preset capability governance center to execute each task action in the task planning sequence to obtain the corresponding enterprise task execution results. Therefore, this embodiment adopts a capability governance-oriented planning domain runtime management mechanism for intelligent agent management. That is, to avoid agents possessing a capability set exceeding the requirements of the business scenario, this invention proposes an agent capability governance mechanism centered on the planning domain. This mechanism includes at least the following: each planning domain is only bound to the minimum capability set required to complete the User Case; when an agent switches between different planning domains, its available capability set changes dynamically with the planning domain; capabilities not required by the current planning domain are invisible and inaccessible at runtime. In this way, through the above design, the capability boundary of the agent is always strongly consistent with the business use case, avoiding the problem of a "jack-of-all-trades agent" generating high-risk operations in complex business systems.

[0041] Specifically, based on the target planning domain, the task planning sequence is pruned to obtain a set of safe and executable actions. Then, the agent is used to call the virtual capability set corresponding to the target planning domain based on the capability virtualization layer in the preset capability governance center, so as to execute each task in the action set based on the virtual capability set and obtain the corresponding enterprise task execution result. This can include: using the agent to send a capability call request to the preset capability governance center; the preset capability governance center includes a capability virtualization layer; the capability virtualization layer is used to uniformly encapsulate and abstract enterprise system interfaces, and maintain a registry including atomic action models and system tools bound to each planning domain; the capability call request includes an action identifier, a parameter list, and a session token; the agent only supports accessing the virtual capability set corresponding to the current planning domain during runtime; The virtual capability set supports unified scheduling and audit logging using the capability virtualization layer; it utilizes a preset authentication center to verify whether user roles have action permissions, quota restrictions, and system tool resource availability based on capability call requests. After successful verification, it calls the preset capability governance center to invoke the corresponding system tools or interfaces to monitor the task execution process and collect real-time metrics to obtain enterprise task execution results; after task completion, it calls the preset capability governance center to return the enterprise task execution results to the intelligent agent, allowing the intelligent agent to update the current system business status based on the enterprise task execution results and record the execution traces in the audit log; it uses a preset unsupervised machine learning algorithm to analyze execution patterns based on the audit logs, identifies deviations from expected behavior and performance bottlenecks, and obtains an audit report; the audit report includes compliance scores, efficiency heatmaps, and root cause analysis.

[0042] It is worth mentioning that this application embodiment, focusing on the engineering implementation needs of enterprise-level agents, constructs a system architecture centered on a planning domain. The value of this solution is mainly reflected in: achieving system-level controllability of agent behavior. By uniformly constraining agent behavior through the planning domain, its execution results are predictable and controllable, meeting enterprise-level security requirements. Enabling governable injection of business knowledge. Business knowledge is extracted from the Prompt and managed uniformly in the form of a structured planning domain, supporting versioning and rapid evolution. Significantly reducing the risk of agents in complex business scenarios. Agents can only act within clearly authorized business boundaries, avoiding unpredictable behavior caused by model uncertainty. Deployment to core enterprise systems can be achieved without modifying large models. All governance capabilities are implemented through the system architecture, reducing the engineering costs and risks of enterprises adopting agent technology.

[0043] Furthermore, key technical points of this application's embodiments include: AI-based reasoning planning domain modeling technology based on User Cases, that is, transforming business use cases into a structured business model that agents can reason about, achieving a unified expression of business knowledge and capability boundaries. A planning domain-based agent behavior governance method. Agent decisions and execution are always constrained by the states, actions, and constraints defined in the planning domain, ensuring controllable behavior. An agent capability governance mechanism based on planning domains. By binding minimal capability sets, the problem of uncontrolled enterprise-level agent capabilities is solved. A planning domain-driven deterministic reasoning and execution mechanism. This enables agents to have highly consistent behavioral patterns under the same business use cases, meeting the stability and predictability requirements of enterprise-level systems.

[0044] As can be seen from the above, this embodiment of the application needs to determine a structured target planning domain based on user use cases in user requests and a preset planning domain library. Then, it obtains real-time business data corresponding to the enterprise backend system through the business system interface. The target planning domain is a planning domain that AI can reason about. Secondly, based on the requirements of the target planning domain, it obtains real-time business data corresponding to the target planning domain through the business system interface, and uses an intelligent agent to parse, organize, and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain. Based on the data structure information and semantic representation information, it determines the real-time business data. Then, it uses an intelligent agent to map the real-time business data to the target planning domain to obtain the current system business state, and constructs a structured reasoning context based on the target planning domain and the current system business state. Then, it sends the reasoning context to a preset large language model service to obtain a task planning sequence. Finally, it uses an intelligent agent and a preset capability governance center to execute each task action in the task planning sequence to obtain the corresponding enterprise task execution results. In this way, the efficiency of using intelligent agents to execute enterprise tasks is improved in the process of enterprise task execution based on intelligent agents, thereby improving the user experience.

[0045] Accordingly, see Figure 2 As shown, this application also provides an enterprise task execution device based on an intelligent agent, comprising: The target planning domain determination module 11 is used to determine a structured target planning domain based on user use cases in user requests and a preset planning domain library, and then obtain real-time business data corresponding to the enterprise back-end system through the business system interface; the target planning domain is an AI-reasonable planning domain. The business data determination module 12 is used to obtain real-time business data corresponding to the target planning domain through the business system interface based on the requirements of the target planning domain, and use the intelligent agent to parse, normalize and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, so as to determine the real-time business data based on the data structure information and semantic representation information. The reasoning context determination module 13 is used to map the real-time business data to the target planning domain using an intelligent agent to obtain the current system business state, and to construct a structured reasoning context based on the target planning domain and the current system business state. Then, the reasoning context is sent to a preset large language model service to obtain a task planning sequence. The enterprise task execution result determination module 14 is used to perform behavior trimming on the task planning sequence based on the target planning domain to obtain a safe and executable action set. Then, it uses the intelligent agent and the capability virtualization layer in the preset capability governance center to call the virtual capability set corresponding to the target planning domain, so as to execute each task in the action set based on the virtual capability set and obtain the corresponding enterprise task execution result.

[0046] In some specific embodiments, the business data determination module 11 may specifically include: The user use case identifier parsing unit is used to parse the corresponding user use case identifier from the user request, and perform multi-dimensional matching and retrieval in the preset planning domain library based on the user use case identifier to obtain the matching and retrieval results; the preset planning domain library is a library stored using a graph database structure; the preset planning domain library includes nodes corresponding to each planning domain; the node attributes include business objectives, business state sets, action models, and constraint rules; all the nodes are connected by edges; The semantic similarity determination unit is used to determine the semantic similarity between the user use case identifier and each node using a graph traversal algorithm and based on the user role attributes and the real-time system context, and to select the nodes with the highest similarity among the semantic similarities using a preset historical execution success rate as a candidate set. The planning domain acquisition unit is used to adaptively calibrate the threshold of state variables and priority of constraint rules corresponding to the candidate set by using a preset dynamic adjustment module and based on real-time business environment data and the matching retrieval results, so as to obtain a structured target planning domain.

[0047] In some specific embodiments, the inference context determination module 13 may specifically include: The business system interface construction unit is used to determine the requirements corresponding to the target planning domain, and to construct a business system interface with a distributed microservice architecture, data cleaning and normalization pipeline. Then, it uses an asynchronous message queue to parse the first business data to be processed, which includes several data change events. The heterogeneous business data acquisition unit is used to perform noise reduction, deduplication and format verification operations on the first business data to be processed in sequence using the business system interface and based on preset data quality rules and the requirements, so as to obtain heterogeneous business data. The business data snapshot acquisition unit is used to organize the heterogeneous business data to obtain structured second business data to be processed, and to use a real-time stream processing engine to perform window aggregation, state calculation and semantic transformation on the second business data to be processed to obtain a real-time business data snapshot; the real-time business data snapshot includes the original data value and business indicators; The data delivery unit is used to send the real-time business data snapshot to the intelligent agent through a preset interface using the business system interface, so as to use the intelligent agent to determine the real-time business data based on the real-time business data snapshot.

[0048] In some specific embodiments, the inference context determination module 13 may specifically include: The mapping relationship matrix determination unit is used to determine the mapping relationship matrix by utilizing the integrated state mapping engine and deep learning model in the intelligent agent, and based on the preset knowledge graph, the real-time business data and the state variables of the target planning domain. The system business status determination unit is used to determine the current system business status corresponding to each of the state variables by using a preset state mapping engine and based on the mapping relationship matrix, and then use a preset inference context construction module to extract the business objectives, action model subsets and constraint rules corresponding to the target planning domain by using the mapping relationship matrix. The reasoning context determination subunit is used to generate a structured reasoning context based on the business objective, the action model subset and the constraint rules, similar case features in the preset historical execution library, the target planning domain and the current system business state; the reasoning context includes a time sequence marker and a priority label.

[0049] In some specific embodiments, the inference context determination module 13 may specifically include: The inference context delivery unit is used to send the inference context to a preset large language model service through the agent and the API gateway; the API gateway is used for data flow control and anomaly detection. The reasoning context processing unit is used to process the reasoning context using the domain-specific language model in the preset large language model service and based on the preset prompt word project to obtain the initial task planning sequence; The task planning sequence merging unit is used to merge, parallelize, and perform resource dependency analysis on the initial task planning sequence using a preset task planning sequence post-processing center and a preset graph optimization algorithm to obtain the target task planning sequence. The target task planning sequence is in the form of a directed acyclic graph. The nodes in the target task planning sequence are atomic actions, and the edges are state transition conditions.

[0050] In some specific embodiments, the agent-based enterprise task execution device may further include: The verification knowledge base construction unit is used to load the atomic action model set, the state transition rule set, and the constraint rule set from the target planning domain using the rule engine, so as to construct a pruning rule base based on the atomic action model set, the state transition rule set, and the constraint rule set; The task action conversion unit is used to convert each task action in the task planning sequence into a time-series logic formula, and to use the pruning rule base to verify whether the task planning sequence meets the safety and liveness attributes. The task action judgment unit is used to determine, if the verification is successful, whether each task action in the task planning sequence belongs to the set of atomic action models corresponding to the target planning domain using the planning verifier. If it belongs, it determines whether the state before and after the execution of each task action follows the state transition rules of the target planning domain. If it does, it determines whether the execution state corresponding to each task action satisfies the set of constraint rules of the target planning domain. The signature certificate generation unit is used to set each of the task planning sequences as a set of secure and executable actions if the conditions are met, and to generate a signature certificate corresponding to the task planning sequence using the planning verifier. If the conditions are not met, a dynamic correction sub-process is triggered, and a correction suggestion is generated based on the rule engine. Then, the suggestion and error code are returned to the agent through the feedback interface to drive the iterative optimization of the task planning sequence.

[0051] In some specific embodiments, the enterprise task execution result determination module 14 may specifically include: A capability invocation request sending unit is used to send a capability invocation request to a preset capability governance center using the intelligent agent; the preset capability governance center includes a capability virtualization layer; the capability virtualization layer is used to uniformly encapsulate and abstract enterprise system interfaces, and maintain a registry including atomic action models and system tools bound to each planning domain; the capability invocation request includes an action identifier, a parameter list, and a session token; the intelligent agent only supports accessing the virtual capability set corresponding to the current planning domain during runtime; the virtual capability set supports unified scheduling and auditing records using the capability virtualization layer; The Enterprise Task Execution Result Determination Subunit is used to verify whether the user role has the right to act, whether there are quota restrictions, and the availability of system tool resources based on the preset authentication center and the capability call request. After the verification is passed, the preset capability governance center is called to call the corresponding system tools or interfaces to monitor the execution process of the task and collect real-time indicators to obtain the enterprise task execution result. The task execution result return unit is used to call the preset capability governance center to return the enterprise task execution result to the intelligent agent after the task is completed, so that the intelligent agent can update the current system business status based on the enterprise task execution result and record the execution trace to the audit log; The audit report generation unit is used to utilize a preset unsupervised machine learning algorithm and analyze the execution pattern based on the audit logs, and identify deviations from expected behavior and performance bottlenecks to obtain an audit report; the audit report includes a compliance score, an efficiency heatmap, and root cause analysis.

[0052] Furthermore, embodiments of this application also disclose an electronic device, Figure 3 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the agent-based enterprise task execution method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0053] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0054] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0055] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the agent-based enterprise task execution method disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0056] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned agent-based enterprise task execution method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0057] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0058] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0059] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0060] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0061] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for executing enterprise tasks based on intelligent agents, characterized in that, include: Based on user use cases in user requests and a preset planning domain library, a structured target planning domain is determined, and then real-time business data corresponding to the enterprise back-end system is obtained through the business system interface; the target planning domain is an AI-reasonable planning domain. Based on the requirements of the target planning domain, real-time business data corresponding to the target planning domain is obtained through the business system interface. The intelligent agent is used to parse, normalize and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, so as to determine the real-time business data based on the data structure information and semantic representation information. The real-time business data is mapped to the target planning domain using an intelligent agent to obtain the current system business state. A structured inference context is then constructed based on the target planning domain and the current system business state. The inference context is then sent to a preset large language model service to obtain a task planning sequence. Based on the target planning domain, the task planning sequence is pruned to obtain a set of safe and executable actions. Then, the agent is used to call the set of virtual capabilities corresponding to the target planning domain based on the capability virtualization layer in the preset capability governance center, so as to execute each task in the set of actions based on the set of virtual capabilities and obtain the corresponding enterprise task execution results.

2. The enterprise task execution method based on intelligent agents according to claim 1, characterized in that, The process of determining the structured target planning domain based on user use cases in user requests and a preset planning domain library includes: The corresponding user use case identifier is parsed from the user request, and a multi-dimensional matching search is performed on the user use case identifier in the preset planning domain library to obtain the matching search results; the preset planning domain library is a library stored using a graph database structure; the preset planning domain library includes nodes corresponding to each planning domain; the node attributes include business objectives, business state sets, action models, and constraint rules; the nodes are connected by edges; The semantic similarity between the user use case identifier and each node is determined by using a graph traversal algorithm and based on user role attributes and real-time system context. The nodes with the highest similarity among the semantic similarities are selected as a candidate set using a preset historical execution success rate. By using a preset dynamic adjustment module and based on real-time business environment data and the matching retrieval results, the threshold of state variables and priority of constraint rules corresponding to the candidate set are adaptively calibrated to obtain a structured target planning domain.

3. The enterprise task execution method based on intelligent agents according to claim 1, characterized in that, The process involves obtaining real-time business data corresponding to the target planning domain through a business system interface based on the requirements of the target planning domain, and using the intelligent agent to parse, normalize, and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain. The real-time business data is then determined based on the data structure information and semantic representation information, including: Determine the requirements corresponding to the target planning domain, and build a business system interface with a distributed microservice architecture, data cleaning and normalization pipeline. Then, use an asynchronous message queue to parse the first business data to be processed, which includes several data change events. Using the business system interface and based on preset data quality rules and the requirements, the first business data to be processed is sequentially subjected to noise reduction, deduplication and format verification operations to obtain heterogeneous business data. The heterogeneous business data is regularized to obtain structured second business data to be processed. The second business data to be processed is then subjected to window aggregation, state calculation and semantic transformation using a real-time stream processing engine to obtain a real-time business data snapshot. The real-time business data snapshot includes the original data values ​​and business indicators. The real-time business data snapshot is sent to the intelligent agent through a preset interface using the business system interface, so that the intelligent agent can determine the real-time business data based on the real-time business data snapshot.

4. The enterprise task execution method based on intelligent agents according to claim 1, characterized in that, The step of using an intelligent agent to map the real-time business data to the target planning domain to obtain the current system business state, and constructing a structured reasoning context based on the target planning domain and the current system business state, includes: The mapping relationship matrix is ​​determined by utilizing the integrated state mapping engine and deep learning model in the intelligent agent, and based on the preset knowledge graph, the real-time business data and the state variables of the target planning domain. The current system business state corresponding to each state variable is determined by using a preset state mapping engine and based on the mapping relationship matrix. Then, the business objectives, action model subsets and constraint rules corresponding to the target planning domain are extracted by using a preset inference context construction module and based on the mapping relationship matrix. A structured reasoning context is generated based on the business objective, the action model subset and the constraint rules, similar case features in the preset historical execution library, the target planning domain and the current system business state; the reasoning context includes time sequence markers and priority labels.

5. The enterprise task execution method based on intelligent agents according to claim 1, characterized in that, Sending the inference context to a preset large language model service to obtain a task planning sequence includes: The intelligent agent is used to send the inference context to a preset large language model service via an API gateway; the API gateway is used for data flow control and anomaly detection. The inference context is processed using the domain-specific language model in the preset large language model service and based on the preset prompt word project to obtain the initial task planning sequence; The initial task planning sequence is merged, parallelized, and resource-dependently analyzed using a pre-defined task planning sequence post-processing center and a pre-defined graph optimization algorithm to obtain the target task planning sequence. The target task planning sequence is in the form of a directed acyclic graph. The nodes in the target task planning sequence are atomic actions, and the edges are state transition conditions.

6. The enterprise task execution method based on intelligent agents according to claim 1, characterized in that, The step of pruning the task planning sequence based on the target planning domain to obtain a safe and executable set of actions includes: A set of atomic action models, a set of state transition rules, and a set of constraint rules are loaded from the target planning domain using a rule engine, so as to construct a pruning rule library based on the set of atomic action models, the set of state transition rules, and the set of constraint rules; Each task action in the task planning sequence is converted into a temporal logic formula, and the pruning rule base is used to verify whether the task planning sequence satisfies the safety and liveness attributes. If the verification passes, the planning verifier is used to determine whether each task action in the task planning sequence belongs to the set of atomic action models corresponding to the target planning domain. If it does, it is determined whether the state before and after the execution of each task action follows the state transition rules of the target planning domain. If it does, it is determined whether the execution state corresponding to each task action satisfies the set of constraint rules of the target planning domain. If satisfied, each task planning sequence is set as a set of secure and executable actions, and a signature certificate corresponding to the task planning sequence is generated using the planning verifier. If not satisfied, a dynamic correction sub-process is triggered, and correction suggestions are generated based on the rule engine. Then, the suggestions and error codes are returned to the agent through the feedback interface to drive the iterative optimization of the task planning sequence.

7. The enterprise task execution method based on intelligent agents according to any one of claims 1 to 6, characterized in that, The process involves pruning the task planning sequence based on the target planning domain to obtain a safe and executable action set. Then, the agent, using the capability virtualization layer in a preset capability governance center, invokes a virtual capability set corresponding to the target planning domain to execute each task in the action set based on the virtual capability set, thereby obtaining the corresponding enterprise task execution result, including: The intelligent agent sends a capability invocation request to a preset capability governance center; the preset capability governance center includes a capability virtualization layer; the capability virtualization layer is used to uniformly encapsulate and abstract enterprise system interfaces, and maintain a registry including atomic action models and system tools bound to each planning domain; the capability invocation request includes an action identifier, a parameter list, and a session token; the intelligent agent only supports accessing the virtual capability set corresponding to the current planning domain during runtime; the virtual capability set supports unified scheduling and auditing records using the capability virtualization layer; The system utilizes a preset authentication center and verifies whether the user role has the authority to act, whether there are quota restrictions, and the availability of system tool resources based on the capability call request. After the verification is passed, the preset capability governance center is invoked to call the corresponding system tools or interfaces to monitor the execution process of the task and collect real-time indicators to obtain the enterprise task execution results. After the task is completed, the preset capability governance center is invoked to return the enterprise task execution result to the intelligent agent, so that the intelligent agent can update the current system business status based on the enterprise task execution result and record the execution trace to the audit log; An audit report is generated by using a pre-defined unsupervised machine learning algorithm and analyzing the execution patterns based on the audit logs, identifying deviations from expected behavior and performance bottlenecks; the audit report includes a compliance score, an efficiency heatmap, and root cause analysis.

8. An enterprise task execution device based on intelligent agents, characterized in that, include: The target planning domain determination module is used to determine a structured target planning domain based on user use cases in user requests and a preset planning domain library, and then obtain real-time business data corresponding to the enterprise back-end system through the business system interface; the target planning domain is an AI-reasonable planning domain. The business data determination module is used to obtain real-time business data corresponding to the target planning domain through the business system interface based on the requirements of the target planning domain, and use the intelligent agent to parse, normalize and semantically transform the real-time business data to obtain data structure information and semantic representation information that conform to the definition of the target planning domain, so as to determine the real-time business data based on the data structure information and semantic representation information. The inference context determination module is used to map the real-time business data to the target planning domain using an intelligent agent to obtain the current system business state, and to construct a structured inference context based on the target planning domain and the current system business state. Then, the inference context is sent to a preset large language model service to obtain a task planning sequence. The enterprise task execution result determination module is used to perform behavior trimming on the task planning sequence based on the target planning domain to obtain a safe and executable action set. Then, it uses the intelligent agent and the capability virtualization layer in the preset capability governance center to call the virtual capability set corresponding to the target planning domain, so as to execute each task in the action set based on the virtual capability set and obtain the corresponding enterprise task execution result.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the agent-based enterprise task execution method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the agent-based enterprise task execution method as described in any one of claims 1 to 7.

Images