Real AI products, dissected
7 case studies. Each deep-dive covers: architecture (with Mermaid diagram), key components with citations, request flow step-by-step, published scale numbers, trade-offs the team explicitly accepted, failure modes seen in production, what to copy and what NOT to copy.
Cursor· Anysphere
How Cursor beats Copilot at code completion — architecture, latency, cost.
Perplexity· Perplexity AI
Search + LLM synthesis + citations, at scale — the answer-engine pattern.
Copilot· GitHub
The first mass-market AI product — architecture, scale, telemetry, evolution.
Claude Code· Anthropic
The agentic coding CLI — sandboxed tool use, permission model, session state.
NotebookLM· Google
RAG-first personal knowledge assistant — audio generation, citation UX.
Notion AI· Notion
AI features embedded in a product surface — the design lessons.
ChatGPT· OpenAI
The reference AI product — scale, cost, safety, iteration.