// topic 03
Token Efficiency
Getting great results at lower cost — context budgeting, prompt caching, model selection, and measuring the tradeoffs between quality, latency, and spend.
Overview
Placeholder — why token efficiency is a design constraint, not an afterthought, once a system runs at real usage volume.
Techniques
| Technique | Lever |
|---|---|
| Prompt caching | Reuse unchanged context across calls |
| Right-sized model routing | Cheap model for easy steps, strong model for hard ones |
| Progressive disclosure in skills | Only load detail when the task needs it |
| Summarization / compaction | Compress history instead of carrying it verbatim |
Placeholder — elaborate each technique with real examples once learned.
Cost Math
Placeholder — worked example comparing cost of a naive vs. optimized approach.
echo "cost = (input_tokens + output_tokens) * rate"
placeholder — swap in a real before/after calculation
Running notes
Placeholder — day-by-day notes go here.