Cursor has released the latest test results of its upgraded AI agent cluster. Under the conditions of only providing the official SQLite documentation, prohibiting access to source code and the internet, the new generation cluster rebuilt SQLite using Rust, and all model configurations achieved a 100% pass rate in the test suite, while the previous version cluster experienced a significant drop in efficiency due to numerous merge conflicts.
With the release of Cursor3, the agent cluster has evolved from a research project into a core product, allowing developers to run multiple AI agents in parallel to collaboratively complete complex development tasks. The new architecture adopts a "planning agent + execution agent" model, where cutting-edge models handle task decomposition and design decisions, while lower-cost models perform specific coding, reducing context burden and improving collaboration efficiency.
Cursor stated that this division of labor not only reduces the deviation issues of agents when performing tasks over long periods, but also effectively lowers code conflicts and redundant development, providing a new implementation path for large-scale AI collaborative programming. As AI programming enters the stage of multi-agent collaboration, software development is gradually evolving from single-model assistance to a highly automated agent cluster collaboration model.
Notably, the company behind Cursor, Anysphere, has recently been acquired by SpaceX, owned by Elon Musk, for $60 billion, further highlighting the strategic value of AI programming tools in the industry.
