Context orchestration: what it is & how it works

Your LLM application works fine in a demo. You ship it to production, and it starts hallucinating on stale data, looping through the same tool calls, and burning through tokens in retry cycles. The model itself is probably fine.