AI agent benchmarks are broken

I don't think "Benchmarks" are the right way to analyze AI-related processes, which is probably similar to the complexity surrounding human intelligence measurements and how well each human can handle real-world problems.