Entrepreneurial Thought Leaders (ETL) Brendan Foody (Mercor) - Agentic Data and the Future of AI
Apr 15, 2026
Brendan Foody, co-founder of Mercor who built a leading provider of agentic data for frontier AI labs. He discusses how agentic data and benchmarks were identified and built, how AI will reshape jobs and create new roles, and practical tactics for sourcing expert evaluators and measuring professional capabilities.
AI Snips
Chapters
Transcript
Episode notes
Rubrics Are The New Training Data
- Agentic data uses rubrics and unit tests as reward functions and benchmarks rather than simple supervised labels.
- These rubrics let labs both evaluate models and train them via reinforcement learning on real-world success criteria.
Apex Benchmarks Professional Capabilities
- Mercor's Apex benchmarks measure economically valuable professional tasks across domains like banking, law, and software.
- Labs and enterprises adopt Apex to choose models that perform on job-relevant distributions, not just academic tests.
Productivity Gains Create New Work
- Productivity gains from AI will expand demand, not eliminate work, unless we automate 100 percent of tasks.
- Historical productivity increases (20x over 225 years) created more jobs and new sectors instead of mass unemployment.