George Cameron of Artificial Analysis warned that running multi-step AI agents without prompt caching increases operating costs tenfold. Input tokens and cache hit rates now represent the largest cost drivers in agent deployments.
FireBrief
FireBrief