In the era of Agents, where AI systems become ever more powerful and general purpose, evaluation becomes more expensive. This talk introduces a scalable, effective and cost-conscious approach: using large language models to power evaluation. While it might sound counter intuitive, the talk will explore how the latest LLM-as-a-Judge techniques allow for reliable evaluation. Whether you're deploying AI in production or benchmarking research models, this session will offer practical insights and strategies for assessing AI—without breaking your budget.