TA-Agent: Tool Augmented Based on Iterative Ideas in Software Engineering

Humans have long been pursuing Artificial Intelligence (AI) that equals or surpasses the human level, and AI agents are considered a promising vehicle for achieving this goal. Artificially intelligent agents are artificial entities that can perceive the environment, make decisions, and take actions. Agents can solve practical tasks based on existing tools. However, the current tools-based Agents suffer from the drawbacks of incomplete loading of tools libraries and redundancy in tools creation. To address these two issues, we introduce the Tools Augment Agent (TA-Agent) based on the iterative thinking of software engineering. Our automation paradigm uses LLM-based agents to achieve enhancement automation through reasoning and acting recall tools. It is designed to enhance existing tools according to human task requirements without creating new tools as much as possible and to continuously optimize and improve the quality of existing tools through iterations. Finally, the tools-augmented task construction and execution processes are explained in detail through experiments, proving the feasibility of TA-Agent and revealing the possibility of a new paradigm of Agent-driven automation. The code is available at https://github.com/Taurus9527/Tool_Augment.