Vibe Check at Replay '26: Shabnam Emdadi on Expensive LLM Retries

Shabnam Emdadi draws parallels between AI agentic frameworks and distributed systems, noting that challenges like idempotent retries, state management, and multi-agent coordination are problems the distributed systems world has already solved. A key practical takeaway for developers putting AI into production: LLM retries are expensive unlike typical HTTP calls, so retry policies and fallback strategies need to be designed much more thoughtfully to avoid unnecessary costs.