10 Ways to Stop Wasting AI Tokens

1. Start a new chat when the goal changes Old context pollutes new work and consumes more tokens. 2. Give the AI a brief, not a brain dump Goal. Constraints. Input. Output. 3. Upload only the files that matter “Analyze my business” with 19 docs is how you reach your Claude limits. 4. Use summaries as handoffs End long chats with: “Summarize the current state, decisions, open questions, and next steps.” Then restart. 5. Use retrieval instead of dumping everything Search first. Feed only relevant new chunks. 6. Trim tool outputs Logs, JSON, HTML, transcripts, scrape results: clean them before sending. 7. Ask for deltas “Only show what changed.” “Only return the patch.” “Only list the 5 highest-impact fixes.” 8. Cache repeated context If your app sends the same system prompt, company docs, or policy text every time, cache it. 9. Make memory boring and short Persistent instructions should be compact. If your memory file becomes a novel, it will be expensive. 10. STOP always using the most expensive model! Small model for simple tasks. Big model for hard judgment.