Prime Intellect's Prime Agent used a 2,209-agent swarm to rewrite itself in Rust: time-to-type 738ms→56ms, 80%+ less memory

Prime Intellect rewrote Prime Agent, its open-source long-running coding agent harness, from TypeScript to Rust, with the agent doing most of the work. A root orchestrator that wrote no product code split the port into topologically ordered tasks; each passed through a planner, an implementer in its own worktree, an adversarial reviewer on a different model, and a verifier running parity checks in a fresh Prime Sandbox. Humans mainly built objective verification: terminal-frame TUI diffs, transcript and model-request diffs, daemon protocol checks and a per-component feature audit. In total 2,209 agents exchanged 16,758 messages and used 228.7B GLM-5.3 tokens over about two weeks, followed by a three-day hillclimbing loop that merged 69+ performance changes. Vendor-measured results: cold time-to-type 737.8ms to 55.8ms, first paint 722.8ms to 23.6ms, post-startup memory 607.4MB to 106.0MB, install size 172.1MB to 59.6MB; Prime Intellect cautions that cross-harness comparisons lack a common benchmark. The codebase is now nine crates with per-session worker isolation, native Windows (beta) and Homebrew installs.