1.5The Architecture of This Book

This book builds your understanding layer by layer:

+---------------------------------------------------------------------+
|                    CAPSTONE PROJECT (Part VI)                       |
|            Complete multi-agent testing system                      |
+---------------------------------------------------------------------+
|  DETERMINISM & ECONOMICS (Part V)  |  GUARDRAILS & SAFETY (Part IV) |
|  Spectrum, protocols, cost         |  Harnesses, timeouts, safety   |
+------------------------------------+--------------------------------+
|                  CASE STUDIES (Part III)                            |
|            OpenObserve council -- real-world validation             |
+---------------------------------------------------------------------+
|              MULTI-AGENT SYSTEMS (Part II)                          |
|         Orchestrator, Swarm, Critic-Actor patterns                  |
+---------------------------------------------------------------------+
|                 REACT FOUNDATION (Part I)                           |
|          Core loop, commands, production implementation             |
+---------------------------------------------------------------------+

Each layer depends on the ones below it. Master the ReAct loop first; everything else builds on it.

Key Takeaways -- Chapter 1

1. Traditional test scripts are imperative and cannot adapt to unexpected situations.
2. Agentic testing uses a Reason+Act loop that observes, decides, acts, and evaluates.
3. Agents provide diagnostic information, not just pass/fail results.
4. Agents are not a replacement for scripts -- they complement them for different use cases.
5. The ReAct pattern is the foundation for all agentic testing architectures.

Common Mistakes -- Chapter 1

• Mistake: Replacing all scripts with agents. Agents are expensive and non-deterministic. Keep deterministic scripts for CI gates and stable regression tests.
• Mistake: Expecting agents to be faster than scripts. Agent tests take 5-15 seconds due to LLM latency. Scripts take 1-3 seconds.
• Mistake: Not setting guardrails. Without step limits, token budgets, and timeouts, agents can run indefinitely and consume unbounded resources.

Exercises -- Chapter 1

Exercise 1.1 (E): List five scenarios in your current project where a test script has failed due to an unexpected UI change. For each, describe how an agent might have handled it differently.

Exercise 1.2 (M): Take an existing Selenium or Playwright test from your project and rewrite the test objective as a single natural-language sentence. What information would an agent need to achieve this objective?

Exercise 1.3 (H): Create a decision matrix for your test suite. Categorize each test as "keep as script," "convert to agent," or "add agent as supplement." Justify each decision based on the factors in Section 1.4.