Developers at HD Research successfully rewrote the Git version control system in the Zig programming language by orchestrating a swarm of autonomous AI coding agents within virtual machines.

Key Points

• The project utilized the Vers CLI to manage infrastructure, injecting environment variables into VMs to enable autonomous agent orchestration.
• Agents operated in infinite loops, performing sync, build, and push operations while utilizing a "succinct mode" to reduce token consumption.
• The resulting "ziggit" software achieved significant performance gains, including up to 100x speedups in specific Git operations compared to the original C-based implementation.
• The Zig-based implementation produced a WebAssembly binary 5.4x smaller than the existing wasm-git project.
• The development process consumed approximately 13 billion tokens over one week, effectively automating months of manual developer labor.

Why it Matters

This project demonstrates that autonomous agent swarms can handle complex, large-scale engineering tasks when managed by a human-steered "top-level" agent. By providing clear architectural goals and iterative oversight, developers can leverage AI to perform significant refactoring and performance optimization that would be prohibitively time-consuming for human teams.