
Making Turborepo 96% faster with agents, sandboxes, and humans
3/30/2026
Introduced significant performance improvements to Turborepo, resulting in up to 96% faster task graph computation. This was achieved through: 1. AI agent-assisted code analysis leading to optimizations like hashing by reference and switching to `xxhash-rust`. 2. Replacing an inefficient Floyd-Warshall algorithm with a multi-source DFS. 3. Developing a new `turborepo-profile-md` crate to generate LLM-friendly CPU profiles in Markdown format, improving the analysis of performance bottlenecks for both humans and agents. This new format includes sorted hot functions and call trees, making profiling data more accessible and actionable.





































