—AI Implementation
Production AI Agent Architecture: 5 Reusable Patterns from DoorDash & Shippy for Shipping Agents to Production
Most enterprise agents die between the demo and production. Two fresh first-party cases — DoorDash's "Ask DoorDash" assistant and the Allen Institute's Shippy — show the gap isn't which model you picked. It's the five engineering things around the model: deterministic tools, an MCP layer that decouples orchestration from business logic, layered memory, evaluating the agent (not the model), and an FDE-style split between domain and platform teams. Here's the reference architecture, with the numbers.
