One screen for one run-time feature. Fill each field in a plain sentence. When it is full, paste it into the "Where AI fits" section of your Build Plan.
The job. One feature, in one sentence, and the run-time role it plays (it generates content for the user, or it works behind the scenes).
The model. Which model and tier, and the one reason it fits: capability, cost, speed, context window, or privacy.
The prompt. Where the prompt lives in your code, and its rules: what it must never do, and the format you need back.
The output. The shape your code expects (the fields and their types), and the value checks you run before acting on it.
The context. How the model gets your facts: stuffed into the prompt, retrieved, or none needed.
The lever. Which rung of the customization ladder you stopped at (instructions, context, tools, fine-tuning), and why you did not climb higher.
The fallback. The spend cap, and what the feature does when the model is slow, down, or clearly wrong.
A field you cannot fill is a decision still open. The chapter that settles each one:
- The job — Where AI fits in your product
- The model — Choose a model you can live with
- The prompt — Prompting is engineering, not wording
- The output — Get output you can build on
- The context — Give the model the facts it wasn't trained on
- The lever — The customization ladder
- The fallback — Wire the model into your build
Keep it to one screen. If it runs longer, the feature is doing more than one job.