CLAUDE IN PRODUCTION

An Architect's Field Guide to Building Real Systems with Claude

Stop treating the model like a chat box. Start treating it like an engine.

Almost everyone meets Claude the same way: a text box, a question, an answer. That is a fine place to start and a terrible place to stop. The moment you try to build something that runs without you watching, an agent that audits fifty thousand records overnight, or routes a customer to a human the instant they ask, you discover the conversation was never the hard part. The engineering around the conversation is. Claude in Production is the field guide to that engineering, and a complete, plain-English companion for the Claude Certified Architect Foundations (CCA-F) exam.

Built around the five official exam domains, the book turns each one into short, readable chapters with a custom diagram, a reusable method, and a hands-on exercise:

• Agentic Architecture & Orchestration. How the agentic loop really works, why Claude never stops at random, and how to branch your application on stop_reason. The hub-and-spoke orchestrator-worker pattern, Need-to-Know context sandboxing, and lifecycle hooks that guard the pipeline on both sides of the model.
• Claude Code Configuration & Workflows. The three-level CLAUDE.md hierarchy and which level wins, why team rules belong in the repo and not on your laptop, custom slash commands and packaged skills, and headless CI/CD automation with the -p flag.
• Prompt Engineering & Structured Output. Why prompt-based JSON eventually crashes your parser and how native tool_use schemas constrain output at the sampling layer, restrictive schemas with enums and bounds, the 2-to-4 example rule, and why your few-shot examples belong in the Gray Zone with explicit reasoning blocks.
• Tool Design & MCP Integration. The Model Context Protocol as a secure boundary gate, its three primitives (Prompts, Resources, Tools), the crucial difference between syntax and semantic errors, and self-correcting retry loops built on tool_result with is_error.
• Context Management & Reliability. Defeating the Lost-in-the-Middle effect with hook-layer trimming and the Case Facts Block, the Message Batches API for high-volume asynchronous work, circuit breakers that cap retries, time-anchoring, claim-source mapping, and the non-negotiable human-escalation law.

Every concept is taught through a recurring cast of working engineers wrestling with real failure modes, so the patterns stick. The book closes with a one-page, color-coded cheat sheet covering all five domains, a glossary of key terms, and exercises you can run against Claude as you read.

This is an independent study guide. It explains the concepts and patterns the certification tests; because platform details evolve, it presents API field names, flags, and figures as faithful patterns to understand and points you to the official Anthropic documentation to confirm current specifics before you build or sit the exam.

Whether you are preparing for the CCA-F or simply moving from clever prompts to production systems, this is the guide that gets you there.

The model proposes. Your architecture disposes.