Module 18 Wrap-up: The Framework Strategist

You have learned that "Smarter AI" isn't always the answer—sometimes "Better Structure" is what matters. You understand that AgentCore is the solution for the high-stakes, multi-step, and deterministic worlds of enterprise software where every step must be auditable and controlled.

---

Hands-on Exercise: The Shift

1. The Scenario

You are designing an AI for a "Mortgage Pre-approval" process.

1. Collect user info (AI Chat).
2. Verify SSN and Credit Score (Hardcoded API Call).
3. Calculate risk (AI Reasoning).
4. Generate approval letter (AI Generation).
5. Save PDF to S3 and email user (Hardcoded Action).

2. The Task

Why is this a better fit for AgentCore than a basic agent?

• Think about Step 2: Do we want the AI to "Decide" if it should check the credit score? Or must it be a mandatory step that stops the process if the score is too low?

---

Module 18 Summary

• AgentCore: The framework for controlled, deterministic AI.
• Determinism: Ensuring Step B only follows Step A.
• State Persistence: Handling tasks that outlive a single chat session.
• Control vs. Autonomy: The trade-off every AI architect must navigate.

Coming Up Next...

In Module 19, we look at the Architecture of AgentCore. We will learn about Deterministic Control Flows, State Management, and how to design workflows that can survive a "Failure" and pick up right where they left off.

---

Module 18 Checklist

• I can explain the difference between a deterministic and non-deterministic workflow.
• I understand why AgentCore is better for long-running tasks.
• I have identified a workflow that requires a "Hard Stop" if a condition isn't met.
• I know when to prioritize "Control" over "Autonomy."
• I can describe a "Stateful" conversation versus a "Stateless" one.