Comment by blovescoffee

Naive question, could agents help speed up building code for ROCm parity with CUDA? Outside of code, what are the bottlenecks for reaching parity?