4/Long sessions cost 3x more per message than short ones. Not because Claude gets slower — because it re-reads 80 messages of history before answering message #81. The fix isn't patience. It's /clear + a CONTEXT.md handoff.
🚫 Anti-Patterns: Stop Doing This
• Don't use AI as a Linter: LLMs are too slow/expensive for this.
• Don't Stack Tasks: One task at a time.
• Don't Let Chats Linger: End the session when the task is done.
Your CLAUDE.md is your leverage. Craft it deliberately. 🏁
♾️ The Continuous Learning Loop
Make your next chat smarter. After fixing a bug, ask the AI to reflect: "What advice would you give yourself regarding this mistake?" Extract that lesson and update your CLAUDE.md. Build a repository of past failures to avoid repeating them.