Always-on agents for the holiday season and beyond | Decagon

This post details Decagon's approach to ensuring AI agent reliability and resilience during high-demand periods like the holiday season. It outlines the architectural pillars: scalable multi-tenant architecture with elastic capacity and isolation of real-time tasks; resilient disaster recovery with multi-LLM/voice provider failover and defense in depth; and operational best practices including proactive capacity planning, a code freeze, and 24x7 monitoring. Key technical elements include autoscaling within seconds, tiered storage for latency control, per-tenant throttling, and automatic traffic shedding under adversarial conditions.