Compliance obligations don't attach to models. They attach to systems. Before you can reason about what your agent must do, you need a precise map of what it is made of.
The developer chapters draw on three tools throughout: LangChain for orchestration and chain-of-thought patterns, LangSmith for observability and audit traces, and Deep Agents for human-in-the-loop approval flows and permission enforcement. Where a pattern is framework-agnostic, the code is illustrative. Where a specific API is shown, it is the most direct implementation available.
Every agent system that handles real data, makes real decisions, or takes real actions can be described through six layers. The layers are not theoretical. They correspond directly to where compliance obligations arise, where data flows across trust boundaries, and where systems fail when they fail.