EnterpriseClawBench: Benchmarking Agents from Real Workplace Sessions

EnterpriseClawBench is a newly introduced benchmark for evaluating enterprise agents, derived from actual workplace sessions, comprising 852 reproducible tasks with detailed metadata. The benchmark emphasizes the importance of reporting various performance metrics such as harness-model combinations, artifact delivery, visual quality, cost, runtime, and skill-transfer behavior, rather than a singular performance score; the best model configuration achieved a score of 0.663 using Codex with GPT-5.5. This benchmark is significant for practitioners as it provides a structured evaluation framework tailored to the complexities of enterprise environments, enhancing the development of more effective AI agents.