Enterprise AI benchmarks are broken

DevRev released the first version of its Enterprise AI Agent Benchmark, an open test with dataset, evaluation harness, judging criteria, and results. The platform argues existing benchmarks like TAU-Bench and Agent's Last Exam fail to test real employee work, where large context window handling matters more than reasoning power.