In the current AI-driven landscape, companies are increasingly integrating AI features into their products, often as a marketing strategy, but not all these features are effective or valuable. Good AI features are defined by their ability to offer low-friction user experiences and leverage unique data that only the product can access. This is evident in products like GitHub's Copilot, which excels by providing seamless AI responses, and Notion's AI search, which operates over proprietary data. Effective AI integration requires that a product already delivers value without AI, ensuring that AI features enhance rather than define the user experience. Additionally, AI features that excel usually have access to unique datasets, allowing them to offer insights or automation that other platforms cannot replicate. Examples from Graphite highlight these principles in action, showcasing AI-generated branch names as a feature that reduces user effort while utilizing unique data. As AI technology evolves, these principles may shift, but the current focus remains on reducing friction and utilizing proprietary data effectively.