Skip to slide
217 / 221

CH 18 · Reference Architecture and a Design Checklist · 12/16

Read as one page

Cost & performance

  • Are tasks tiered to the cheapest capable model, with reasoning off for bulk? (Ch 4, 14)
  • Do you reduce iterations (batching, good descriptions, discovery tools) and parallelize independent work? (Ch 14)
  • Do you cache external calls and exploit prompt caching with a stable prefix? (Ch 9, 14)
  • Is usage metered for per-user limits/attribution? (Ch 9, 14)

Finished this idea?

Your learning progress stays in this browser.