Cursor announced what it calls a 99% speed improvement for AI code search through a new local indexing system that pre-processes codebases to help AI agents find relevant files in milliseconds. The company claims this "instant grep" functionality allows their AI coding assistant to narrow down large codebases before running regex queries, dramatically reducing search latency that previously bottlenecked AI-assisted development workflows.
This addresses a real pain point for AI coding tools working with enterprise codebases. When GitHub Copilot or similar assistants need context from thousands of files, search latency becomes the limiting factor — not model inference speed. If Cursor's claims hold up, this could be the difference between AI suggestions that feel responsive versus ones that kill your flow state. The focus on local indexing also sidesteps the privacy concerns that come with uploading entire codebases to external services.
However, Cursor provided surprisingly few technical details about how this indexing actually works. There's no information about memory requirements, indexing time for large repos, or how the system handles code changes. The 99% figure sounds impressive but lacks context — 99% faster than what baseline? Their previous system? Competitors? Raw grep? Without benchmarks against tools like ripgrep or ag, it's hard to evaluate whether this is a genuine breakthrough or clever marketing around incremental improvements.
For developers already using Cursor, this should be a noticeable improvement if you work with large codebases. For teams evaluating AI coding assistants, fast local search is becoming table stakes — but wait for more detailed benchmarks before making this a deciding factor.