US20260099791
2026-04-09
Physics
G06Q10/0633
Generative AI agents are designed to enhance workflow automation by utilizing advanced technology to manage and streamline complex processes. These agents are implemented within Enterprise Resource Planning (ERP) systems, which are integral for managing various business functions such as financials, supply chain operations, and human resources. ERP systems often employ intricate workflows, which can be challenging to navigate due to their complexity. By integrating generative AI, these systems can provide more intuitive guidance, reducing the potential for errors and inefficiencies.
The integration of generative AI involves the use of large language models (LLMs) to generate text descriptions of processes within a workflow. These descriptions help in understanding the tasks and their sequences, making it easier for users to interact with the workflows. The AI system binds a set of tools to the workflow, specifying application programming interfaces (APIs) necessary for performing tasks. This setup enables the creation of autonomous agents capable of executing specific tasks in response to user queries, thereby improving the overall efficiency and accuracy of workflow execution.
Autonomous agents play a crucial role by executing selected tasks within a workflow when prompted by user queries. These agents identify tasks by comparing user queries with the text descriptions generated by the LLMs. This approach not only automates certain tasks but also provides users with step-by-step assistance, context-aware suggestions, and real-time feedback. Such capabilities significantly enhance the user experience by simplifying the navigation of complex workflows and ensuring consistent and accurate execution.
The ERP system architecture supporting generative AI agents includes various platform services such as connectivity, data privacy, and authorization management. A generative AI hub, which may be hosted externally or locally, provides the necessary AI capabilities. This hub includes an embedding model and a large language model, which work together to transform input text into meaningful representations and generate natural language responses. The system also features a workflow assistant for design and runtime phases, allowing administrators to create and deploy autonomous agents, and end users to interact with these agents efficiently.
During the design phase, workflows are compiled using a workflow compiler and a workflow engine. This process involves parsing the workflow's markup language definition and representing each process as a set of nodes. Detailed text descriptions, or context prompts, are generated for each process, providing a comprehensive overview of the tasks and their sequences. These context prompts are crucial for the autonomous agents during the runtime phase, as they offer the necessary contextual information to execute tasks effectively. This structured approach ensures that all processes within a workflow are clearly understood and efficiently managed.