Cursor's agent swarm suggests cheaper models can handle most coding when frontier models plan the work

Cursor tasked its upgraded agent swarm and its predecessor to rebuild SQLite in Rust using only the documentation, without source code or internet access. Every configuration of the new system, which separates planners from workers, scored 100 percent on the test suite. In contrast, the old swarm struggled with merge conflicts it created itself. The results from Cursor's agent swarm indicate that cheaper models can handle most coding tasks when frontier models plan the work.