A model is not a feature you add. It is a component with five properties no other component in your stack has, and almost every strange thing about building AI products is a consequence of one of them. Seven lessons on what those properties are, what they break, which shape of product you are actually building, and when the right answer is to use no model at all.
| # | Lesson | The question it answers |
|---|---|---|
| 01 | The Five Facts | What is structurally different about a model, and what follows from each property? |
| 02 | Deterministic and Probabilistic Products | What breaks in your spec, your tests, your support and your pricing? |
| 03 | The Capability Map | What can models do and not do today, and how do you keep the answer current? |
| 04 | AI Product Shapes | Copilot, agent, autonomous workflow, native app, infrastructure: which are you building? |
| 05 | Feature or Product | Where does value accrue when the model is somebody else’s? |
| 06 | When Not to Use a Model | What is the deterministic alternative, and how do you argue for it? |
| 07 | The 2026 Landscape | What fills each box of the stack, and which parts will have changed by next year? |
Lesson 04 is a prerequisite for Product by Sector. Shape usually determines more about a product than the market it sells into.