Manuals / Prompt Engineering / Chapter 3
A · Structure · beginner · ~30 min · Chapter 3 of 12
Context windows & token economy
Every model has a context limit. Long prompts cost money and attention. Put the most important info first and last — models sometimes lose the middle.
Path progress21%
Step 1 of 2
Budget your context
Estimate: system instructions + examples + user task + expected output. Trim fluff.
Do this now
Take a long doc you paste into chat. Summarize it first, then ask questions on the summary.
Pro tip: For code: paste the relevant function, not the whole repo.
Was this step clear?
Chapter learning outcomes
- Context limits
- Token budgeting
- Lost-in-the-middle
Clear these before you leave
Side quest
Context audit
Review your last 5 prompts. Cut unnecessary context from one.