« LLMs can successfully author CODA kernels » That might speed up progress in this area then