What the journey covered
The book moved from the strategic foundations of GenAI and the agent anatomy, through the interoperability stack and the design-pattern catalog (coordination, explainability, robustness, interaction, agent- and system-level), into advanced adaptation, a maturity roadmap, and hands-on use cases. The throughline is a single idea: pattern-first architecture — building agentic systems from proven, repeatable patterns rather than ad-hoc prompts.
Higher maturity is patterns working together
Reaching a higher level of maturity means composing patterns so each non-functional requirement is met. A worked PTO-request example shows how a single workflow pulls in several patterns at once:
| Requirement | Pattern applied | Why |
|---|---|---|
| Compliance | Instruction Fidelity Auditing | The workflow handles employee-specific data that must stay within policy. |
| Safety | Human-in-the-Loop (Agent Calls Human) | It modifies a database (submits the PTO request) — a human confirms. |
| Robustness | Adaptive Retry | The external HRIS API might fail, so calls retry with backoff. |
Key concept · Pattern-first is strategic budgeting of focus
Adopting patterns deliberately, by level, concentrates limited organizational focus where it produces validated maturity and business value — rather than an unfocused attempt to "build AI." Higher maturity is not one big leap; it is the disciplined composition of the right patterns for the requirements at hand.
From theory to practice
The practitioner's action plan bridges the gap from understanding to skill: master one agentic framework by building something small but real — not another "hello world." A strong first project is an agent that monitors a GitHub repository, built up pattern by pattern.
Start with tool use (Ch. 4)
Give the agent two real tools — e.g. get_latest_issues(repo) and send_email(...) — forcing you to handle real API auth, rate limits, and JSON.
Add a robustness pattern (Ch. 7)
Wrap the flaky API call in Adaptive Retry with exponential backoff so the agent degrades gracefully instead of crashing.
Add an interaction pattern (Ch. 8)
Insert a Human-in-the-Loop check before the agent sends an email, so a person approves consequential actions.
Pick the framework that fits your style — LangGraph for explicit stateful control, CrewAI for the team-of-specialists metaphor — and build a tangible goal that implements the book's patterns. The goal is to become the catalyst for your organization's agentic transformation.
Chapter 16 quiz
Twelve questions on the recap, composing patterns, and the practitioner's action plan. Answer first, then expand Show answer.
Part A · Multiple choice
What single idea is the throughline of the whole book?
- A Always use the biggest model
- B Pattern-first architecture — building from proven patterns, not ad-hoc prompts
- C Avoid multi-agent systems
- D Never involve humans
Show answer
B. Repeatable patterns are the foundation of reliable agentic systems.
In the PTO-request example, which pattern satisfies the safety requirement of modifying a database?
- A Adaptive Retry
- B Human-in-the-Loop (Agent Calls Human)
- C Instruction Fidelity Auditing
- D Blackboard
Show answer
B. A human confirms the consequential write action.
In that example, which pattern handles a possibly-failing external HRIS API?
- A Adaptive Retry
- B Instruction Fidelity Auditing
- C Human-in-the-Loop
- D Formation Control
Show answer
A — Adaptive Retry. It retries the flaky call with backoff for robustness.
What is the recommended first step of the practitioner's action plan?
- A Read three more books
- B Master one agentic framework by building something small but real
- C Buy a GPU cluster
- D Fine-tune a model from scratch
Show answer
B. Move decisively beyond "hello world" and build a tangible project.
In the suggested GitHub-monitor project, what do you add after basic tool use?
- A A robustness pattern (Adaptive Retry with backoff)
- B Nothing
- C A new model
- D A bigger prompt
Show answer
A. Then an interaction pattern (Human-in-the-Loop) before sending email.
Reaching higher agentic maturity is best described as:
- A One big technological leap
- B The disciplined composition of the right patterns for the requirements at hand
- C Removing governance
- D Using a single monolithic agent forever
Show answer
B. Patterns compose to meet compliance, safety, robustness, and more.
Part B · True or false
The action plan advises staying at "hello world" tutorials.
Show answer
False. It urges moving decisively beyond them to build something small but real.
A single workflow can legitimately require several patterns working together.
Show answer
True. The PTO example composes compliance, safety, and robustness patterns.
Pattern-first architecture is a way of budgeting organizational focus.
Show answer
True. It concentrates effort where it yields validated maturity and value.
Building a first project that connects to a live API teaches you to handle auth, rate limits, and JSON.
Show answer
True. Real tools force real-world engineering challenges.
Part C · Short answer
Name the three patterns composed in the PTO-request example and the requirement each meets.
Show answer
Instruction Fidelity Auditing (compliance), Human-in-the-Loop (safety), and Adaptive Retry (robustness).
List the three build steps of the suggested first project.
Show answer
1) Start with tool use, 2) implement a robustness pattern (Adaptive Retry), 3) implement an interaction pattern (Human-in-the-Loop).