Self-Evolving Software Agents Using BDI and LLMs

A new paper introduces self-evolving software agents that combine BDI (Belief-Desire-Intention) reasoning with large language models (LLMs) to autonomously evolve their goals, reasoning, and executable code. The proposed BDI-LLM architecture includes an automated evolution module that works alongside the agent's reasoning loop, enabling it to derive new requirements from experience and generate corresponding design and code updates. A prototype tested in a dynamic multi-agent environment demonstrated that agents can discover new goals and produce executable behaviors with minimal prior knowledge. The study highlights both the potential and current limitations of LLM-driven evolution, particularly regarding behavioral inheritance and stability.