A 24% Success Rate for AI Agents – Is That Acceptable?

A major benchmark study testing AI agents on real-world professional tasks from investment banking, consulting, and law found they succeed only 24% of the time on first attempts, though performance improves to 36-40% with multiple tries. While agents struggle with execution and maintaining consistency across complex workflows, they can still provide value by handling focused tasks like data extraction and initial analysis when integrated into human-supervised processes.