Yoyo-Evolve puts autonomous code evolution on public display
A Rust-based coding agent is running an unattended self-improvement loop: it reads its own code and journal, processes external GitHub issues, ships changes when tests pass, and reverts on failure. The project has grown from ~200 to 1,500+ lines in four days, with public logs showing both real progress and realistic failure patterns.
This is a compelling real-world stress test for agentic coding because it exposes long-horizon behavior, not just one-shot demos.
- –Public commit history and journals make agent progress auditable, which is rare for autonomous coding experiments.
- –Self-filed issues and deferred tasks show early planning behavior, including prioritization failures like repeated streaming procrastination.
- –The loop design (test-gated commits plus rollback) is a practical safety pattern for autonomous code changes.
DISCOVERED
83d ago
2026-03-05
PUBLISHED
83d ago
2026-03-04
RELEVANCE
AUTHOR
liyuanhao