Asana cut a Codex agent's cost 4x to $0.47 per run with an 89% prompt cache hit rate
daniel_mac8 · x · 2026-10-11
Asana cut the cost of its Codex-built browser agent on GPT-6.1 Sol by 4x ($1.97 → $0.47 per run) with one move: keeping the agent's history stable so the prompt cache hits 89% of the time.
danielmac8 distilled four habits you can adopt today to maximize cache hit rate:
- Keep sessions append-only: don't rewrite earlier context, and stay in one long session instead of restarting.
- Don't switch model, reasoning effort (except GPT-6 Astra), or tools mid-session — any change alters the start of the request and breaks the cache.
- Treat /compact as a cache reset: it rewrites earlier context, so compact deliberately, not by habit.
- Measure it: run codex exec -- "<task>" and divide cachedinputtokens by inputtokens.
The approach works well for individuals and compounds for teams as token usage scales.
Related event: Asana Cuts Codex Agent Costs via Stable Prompt Caching(3 posts)→
More from coding & agent
- Python vs Go vs Rust for production agentic systems: a six-dimension practical evaluation — mostly_deterministic · 2026-10-11
- Agent Use Cases: a free site scanning the web daily for 1,000+ real AI agent use cases — Roger_M_Taylor · 2026-10-11
- Theo: Modern models handle mid-task steering well enough to skip message queueing — pvncher · 2026-10-11
- 1,680 A/B trials show 50-year-old Unix formalisms crush English prompt rules, 98.3% vs 0.8% — RandalSchwartz · 2026-10-11
- Creator tests viral open-source REA to reverse engineer CapCut's hidden tricks — vista8 · 2026-10-11
- Running 426GB MXFP4 DeepSeek on 192GB VRAM to power a 4-sub-agent code review — HankYeomans · 2026-10-11