Persistent Recursive Worlds Enable Autonomous Software Evolution

2026-08-11

AI Summary

This work addresses the challenge that traditional agent-based systems, reliant on persistent sessions, struggle to support long-term software evolution across agent lifecycles. To overcome this limitation, the authors propose the EvoX Genesis framework, which centers on persistent software projects modeled as recursive worlds. Within this paradigm, short-lived agents propose changes along localized paths, and a recursive delegation mechanism drives evolutionary progress, with only accepted changes updating the project’s version history. The framework leverages large language models such as DeepSeek-V4 Flash and GLM-5.2 for code generation. Empirical validation includes the zero-shot construction of a Rust-based C compiler passing c-testsuite (250k lines, costing merely \$44) and the rewriting of 13 MESA modules, achieving 1.55–6.87× performance improvements—demonstrating development continuity and sustained performance under agent replaceability.

Abstract

Complex software systems develop over timescales that exceed the lifespan of any individual coding agent. Most agentic software systems preserve continuity through persistent sessions, memories, managers or shared context. We introduce EvoX Genesis (hereafter, Genesis), which instead makes the software project persistent while allowing local agents to remain finite-lived. Genesis represents software as a persistent recursive world: each local world is situated by an accepted version and a repository path, finite-lived agents propose local changes, recursive delegation moves work across paths, and only accepted consequences advance the persistent version history. We evaluate this organization across formation, continuation and redevelopment. Starting from a repository with no compiler implementation, Genesis used DeepSeek V4 Flash to build a Rust-based C compiler with about 250k tracked lines; the run lasted over 120 hours, archived over 1,000 agent episodes and incurred only US$44 in model-token charges. The compiler passed the complete c-testsuite and most LLVM and Csmith tests. In a separate compiler world generated with GLM 5.2, development continued after repeated agent replacement while retaining full test performance. Genesis also reimplemented 13 MESA modules with over 100k Fortran lines as a Rust workspace with nearly 90k Rust lines; across six numerical workloads, it achieved median speedups of 1.55--6.87x. These results show that long-horizon software development can be organized around a persistent project rather than a persistent agent.

Problem

• autonomous software evolution
• persistent software projects
• long-horizon development
• finite-lived agents
• recursive worlds

Innovation

• persistent recursive world
• autonomous software evolution
• finite-lived agents
• recursive delegation
• long-horizon software development