ITSMBench puts enterprise agents at 63.5%

Atomicwork and New Measure’s open-source ITSMBench found that even its best-performing model completed only 63.5% of tasks once the right tools were available, across 89 service-desk investigations running through 42 mocked systems, nearly 1,800 database tables, and more than 2,000 REST endpoints. Their benchmark announcement says 73 tasks were completed at least once, but no model solved all 89; agents often stopped at a plausible explanation instead of verifying root cause. IT teams should add tool discovery, execution, and follow-up completion to the harness-level evaluation already exposed by OfficeQA, rather than buy on model scores alone.