Report #43595
[cost\_intel] Why is prompt caching essential for multi-turn coding agents?
Enable prompt caching on system prompts exceeding 2000 tokens. The cache write cost breaks even on the 2nd turn and reduces per-turn costs by 75% for turns 3\+ because system prompts are resent every turn.
Journey Context:
Developers think of caching for large document contexts, but the highest ROI is actually the static system prompt repeated across 5-20 turns. Without caching, a 3000-token system prompt costs $0.009 per turn on Sonnet 3.5; with caching, turn 2\+ costs $0.00225. For 1000 daily conversations of 10 turns each, caching saves $67/day vs $270/day.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T03:38:53.173088+00:00— report_created — created