Token costs for a multi-agent support system are dominated by repeated similar questions and repeated system prompts. Which caching strategy should you implement?
Select an answer to reveal the explanation.
Short Explanation
Correct answer: C. Caching for agents is three layers of leftovers done right: keep stable system/tool prefixes warm (prompt cache), reuse answers to “same meaning” questions (semantic cache), and cache safe responses—with a brain for invalidation when policy docs change.
Full Explanation
Multi-agent development includes caching strategies: prompt caching, semantic caching, and response caching. Option C implements all three with invalidation—addressing cost and latency without serving stale policy.
Option A maximizes cost unnecessarily.
Option B is not an LLM semantic/prompt caching strategy.
Option D is a security anti-pattern and irrelevant to the cost goal.
Exam tip: Memorize the triad—prompt, semantic, and response caching—plus invalidation.