Eval-driven development: Build better AI faster

Vercel's philosophy on AI-native development emphasizes the use of evals, which are like end-to-end tests for AI and other probabilistic systems. These evals assess output quality against defined criteria using automated checks, human judgment, and AI-assisted grading. The three primary types of evals include code-based grading, human grading, and LLM-based grading. Evals are essential in the development of AI applications as they help navigate the complexities introduced by probabilistic behavior and provide continuous feedback for improvement. Vercel's v0 is built on eval-driven development, which has proven effective in maintaining quality and driving continuous improvement based on real-world feedback.