Salesforce researchers developed EGI-Bench to assess whether AI agents are ready for enterprise deployment. The framework combines eight benchmarks into six skill dimensions covering over 1,500 test tasks. It evaluates capabilities such as task outcomes, tool actioning, planning, grounding, safety, and engineering. Testing 14 large language models revealed that no single model excels in all areas. The tool is available on GitHub for developers to customize evaluation weights based on specific business needs.
Read the full story at engineering.salesforce.com.
