A Google UX strategist presents frameworks from Google's People and AI Guidebook for designing AI products that align with user intent. Key concepts include human-AI alignment (defining primary goals, sub-goals, underspecification, and optimization trade-offs), shaping user mental models progressively from beginner to expert, providing three types of helpful explanations (proactive/influential features, contrastive/counterfactual, and metric-driven), and managing errors gracefully as trust-building opportunities. The talk uses a hypothetical AI-powered IDE called 'Velocity' to illustrate practical applications, emphasizing that trust must be continuously calibrated rather than treated as a one-time onboarding achievement. Four key takeaways: build continuous translation layers for messy user goals, design adaptive interfaces that progressively reveal AI capabilities, actively design against both overtrust and undertrust, and treat errors as feedback loops rather than failures.