The patterns of AI engineering → API to Architecture

1. The patterns nobody named yet

When traditional software engineering grew up, it accumulated a vocabulary of design patterns. Factories, observers, repositories, circuit breakers. The patterns made it possible for two engineers who hadn't met to discuss a system and understand each other in the same shorthand.

AI engineering hasn't done that yet. The patterns exist — engineers use them every day — but they don't have stable names. One company calls something a router. Another calls the same thing a dispatcher. A third calls it a planner. The lack of shared vocabulary slows everyone down. The patterns are also the things that decide whether your AI feature scales or collapses under load.

2. What API to Architecture covers

[[api-to-architecture]] is the patterns catalog for AI engineering. It names the patterns, defines them precisely, and shows each one with real TypeScript code you can read end to end.

The catalog covers the patterns that show up across production AI systems: retrieval, routing, multi-step workflows, agent loops, guardrails, eval pipelines, feedback loops, the structures that make AI features reliable at scale. Each pattern is presented with the problem it solves, the structure that solves it, the code that implements it, the tradeoffs, and the failure modes.

It's the kind of reference you'll come back to when you're designing a new AI system and need to make architectural calls without reinventing the patterns from scratch.

3. Who should take it next

[[api-to-architecture]] is paid (A$49) and it assumes you've shipped code before. If you're a working engineer designing AI systems — especially at the senior or staff level, where you're making the architectural calls — this is the course.

If you're earlier in your career, take [[vibe-to-spec]] and [[prompt-to-production]] first. Come back to this when you're the one drawing the boxes.