EvoGraph: Hybrid Directed Graph Evolution Toward Software 3.0 (arxiv.org)

Researchers have introduced **EvoGraph**, an innovative framework designed to enable software systems to autonomously evolve their source code, build processes, documentation, and task management. By using a typed directed graph to represent each software artifact, EvoGraph employs specialized small language models (SLMs) to apply learned mutation operators and select the best results based on a multi-objective fitness approach. The framework demonstrated remarkable efficacy by fixing 83% of known security vulnerabilities and achieving a 93% functional equivalence in translating COBOL to Java. Additionally, it maintains documentation freshness in under two minutes, while also reducing latency by 40% and feature lead time by sevenfold compared to traditional methods. The significance of EvoGraph lies in its potential to revolutionize legacy software modernization, addressing common challenges such as performance preservation and integration evolution, all while minimizing computational costs by 90% versus larger models. By facilitating a more adaptive and efficient approach to software evolution, EvoGraph may herald the era of Software 3.0, where systems can continuously adapt and improve while remaining within controlled parameters. This innovation not only enhances productivity but may also set the stage for more intelligent and resilient software infrastructures in the ever-evolving tech landscape.