Reading Paths

The decision it resolves: which chapters, in which order, for the problem you actually have?

Stub: planned for a later release.

By use case and role. "I'm building a chatbot / a RAG app / a batch pipeline / a coding-agent integration / an autonomous agent," "I'm securing one," "I'm a lead deciding whether to," and two paths for the reader who already runs agents all day: "what do I still review?" and "I'm running one unattended tonight." Each path is the chapters, in order, with a sentence on why that order. The conceptual nav stays primary; this is the overlay for the reader who arrives with a problem, not a taxonomy.

It also carries the build path: the chapters sequenced as a build, from a fifty-line loop to a production system (2.5 The Agent Loop onward through tools, structured output, the harness, guardrails, sandboxing, the outer loop, evaluation, compaction, checkpointing, observability, deployment, long-running agents), with the companion repo growing along it and a feature-to-chapter traceability table at the end. A build path, not a build book.