As AI agents take on longer, more complex workflows, maintaining accurate and relevant context becomes increasingly difficult. Agents must retrieve the right information, retain useful memory, and adapt to changing data without relying on outdated assumptions. Poor context management leads to incorrect decisions, increased latency, and unnecessary costs. This panel explores how engineers can design retrieval pipelines, manage agent memory, and maintain data freshness to support reliable, long-running agents in production. Key takeaways: → Designing retrieval pipelines that deliver accurate, relevant, and up-to-date context. → Managing agent memory while preventing context overload and unnecessary costs. → Maintaining data freshness and reliability across long-running, multi-step workflows.