Chapter 4 Declarative Prompts and Type Contracts

Prompt engineering should be a branch of software engineering -- following the fundamental principles of software engineering while developing methods specific to the new paradigm.

This chapter covers two core concepts: Code as Prompt -- using Pydantic models to simultaneously serve as type definitions, semantic instructions, and output validators, where the code itself is the prompt; and Schema as Workflow -- where the arrangement of fields in a Schema defines the LLM's reasoning path, making structure the workflow.

Articles

• From Imperative to Declarative -- From assembly to C, from jQuery to React, from natural language prompts to typed structural declarations -- every leap in abstraction level follows the same structural pattern. LLM prompt engineering is the latest instance of this pattern.
• Code as Prompt -- The three semantic layers of a Pydantic model (type annotations = constraint layer, Field description = semantic layer, Validator = invariant layer) make the code itself a prompt. Define once, enforce in three places. JSON Schema is the machine-readable projection of this model; its expressiveness has boundaries, and beyond those boundaries, runtime validation fills the gap.
• Schema as Workflow -- The arrangement of fields in a Schema defines the LLM's reasoning path. Each field carries one reasoning step, and explicit dependency declarations between fields establish logical connections between reasoning steps. Field-level single responsibility is the precondition for declarative chain-of-thought.

Reading Order

The first article establishes the historical context: the shift from imperative to declarative is a recurring pattern. The second and third articles are the core of this chapter: Code as Prompt and Schema as Workflow, which together form the complete picture of declarative chain-of-thought.