Saving Tokens and Working Longer
How do you reduce token usage in long AI conversations? Start a new conversation when a topic ends and carry over a short summary, let the AI read only the needed parts of files, pre-bundle review material, and cap the answer length.
The longer a conversation runs, the more each reply costs and the blurrier earlier decisions get. Managing length is a skill too.
How to use it
- When a topic is done, start a new conversation. Carry over only the decisions, in a short summary.
- Make the AI read only the parts of files it needs. Letting it explore a whole folder freely can cost millions of tokens at once.
- When asking for review of large material, gather only the content to review into one file beforehand. Do not give just a path and let it search.
- Set the answer length. Reports in 3 to 5 lines, tables only when needed.
- Save frequent instructions in an instruction file or a skill. But a loaded skill keeps taking up space, so use only what you need.
Prompts to copy
Hand off to a new chat
Summarize only what was decided in this conversation, in 10 lines or fewer. - In the order: decisions / reasons / remaining tasks - One line each for approaches tried and dropped - In a form I can paste straight into a new conversation