Back to today's list

Self-Evolving Software Agents

Marco Robol, Paolo Giorgini

Published May 1, 2026
Editorial review6.8
Relevance0.490
Freshness0.000

Why It Matters

What makes this one worth your time

This research could lead to more adaptive and autonomous software agents capable of evolving in response to changing environments, which is valuable for dynamic and complex systems.

Self-evolving software agents combine BDI reasoning with LLMs to autonomously adapt and evolve.

Summary

The paper introduces self-evolving software agents that integrate BDI reasoning with large language models (LLMs) to autonomously evolve their goals, reasoning, and executable code. A prototype is evaluated in a dynamic multi-agent environment, demonstrating the agents' ability to discover new goals and generate executable behaviors with minimal prior knowledge.

Key contributions

  • Introduction of a BDI-LLM architecture for self-evolving agents.
  • Prototype evaluation in a dynamic multi-agent environment.
  • Demonstration of autonomous goal discovery and behavior generation.

Notable insights

  • Combining BDI reasoning with LLMs for autonomous evolution of software agents.
  • Automated evolution module that synthesizes design and code updates from experience.

Possible limitations

  • Not stated in the abstract

Abstract

arXiv:2604.27264v1 Announce Type: cross Abstract: Autonomous agents can adapt their behaviour to changing environments, but remain bound to requirements, goals, and capabilities fixed at design time, preventing genuine software evolution. This paper introduces self-evolving software agents, combining BDI reasoning with LLMs to enable autonomous evolution of goals, reasoning, and executable code. We propose a BDI-LLM architecture in which an automated evolution module operates alongside the agent's reasoning loop, eliciting new requirements from experience and synthesizing corresponding design and code updates. A prototype evaluated in a dynamic multi-agent environment shows that agents can autonomously discover new goals and generate executable behaviours from minimal prior knowledge. The results indicate both the feasibility and current limits of LLM-driven evolution, particularly in terms of behavioural inheritance and stability.