Lesson 2 of 6
The 2026 AI stack
7 min read
Behind a simple 'ask our docs anything' box sits a whole stack of moving parts. What are the layers — and which ones are yours to build?
Four layers under one feature
An AI product isn't one thing — it's a stack. At the bottom sits the model, the raw engine. Above it, data & retrieval feeds the model your knowledge. Above that, orchestration decides what to call and when. And on top, the app — the part the user actually touches. Tap through the layers and see what lives in each.
An AI feature is a stack: model at the base, then data & retrieval, then orchestration, then the app on top.
You mostly live above the model
Here's the punchline: you rarely build the bottom layer. The model comes ready-made, reached through an [API](glossary://api). Your job is almost everything above it — connecting your knowledge with [retrieval](glossary://rag), wiring the orchestration, and building the app. That's why the same handful of foundation models power thousands of very different products.
Layers can be thin or thick. A weekend project might be just 'app + model'. A serious product grows retrieval, orchestration, evaluation, and monitoring as it matures.
The model is shared and bought; the layers above it are where you build your product — and your edge.
The shape of it
- —Model — the foundation engine, reached through an API. You rent it, you don't build it.
- —Data & retrieval — turns your documents into answers the model can use.
- —Orchestration — prompts, chains, agents, routing: the logic that runs the show.
- —App — the interface and product the user actually touches.
Your team switches from one foundation model to another. Which layers of your stack do you mostly keep?
Continue in the app
Take the whole Becoming an AI Engineer course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.