Ben Hylak Outlines Practical Framework For Evaluating AI Agents