Ten unit tests teach AI coding agents to match expert GPU kernels
HAWKEYE grounds coding agents in a minimal taxonomy: one unit test per optimization strategy per architecture, about ten expert-written tests per new accelerator. Scaled with test-time compute, agents then write hardware-aware kernels that match or beat hand-tuned PyTorch and Triton across Ampere, Hopper, Blackwell and MI350, in FP8, NVFP4 and MXFP4. It targets the real bottleneck of the ML-chip explosion: software support, not hardware.