This paper proposes a mechanism called the Action Engine, which leverages a tool-augmented large-scale language model (LLM), to address the challenges of developing applications based on Function as a Service (FaaS), a form of serverless computing. The Action Engine interprets users' natural language queries and automatically generates FaaS workflows, reducing the need for expert knowledge and manual design. It identifies relevant functions in the FaaS repository, manages data dependencies between functions, and executes the generated workflows by injecting user-provided arguments. Furthermore, it addresses the gap in tool-augmented LLM research from the perspective of automatic FaaS workflow generation and systematically evaluates the methodology across four fundamental subprocesses. Experimental results demonstrate that the Action Engine achieves comparable performance with just a few training iterations while remaining platform- and language-independent, mitigating vendor-specific dependencies in workflow generation. Ultimately, the Action Engine enables developers without cloud expertise to create FaaS workflows and can shorten the development cycle of cloud-native applications.
