Beyond the Single Prompt

Simple AI tasks (like "What is 2+2?") can be handled in a single call. However, production-grade tasks (like "Analyze this legal contract, compare it to our internal policy, and generate a risk report") require a sequence of steps.

In the AWS Certified Generative AI Developer – Professional exam, you will be tested on your ability to orchestrate these steps. While you can use code (Python/Lambda) for small chains, AWS Step Functions is the industry standard for building resilient, multi-step AI workflows.

---

1. Why Orchestrate? (The Chain of Thought)

LLMs perform significantly better when they are asked to "Think Step by Step."

• Step 1: Extract key facts.
• Step 2: Analyze facts against a rubric.
• Step 3: Draft the final response.

If you try to do this in one massive prompt, the model is more likely to miss details or hallucinate. By orchestrating distinct calls, you increase Accuracy and Observability.

---

2. AWS Step Functions for AI

AWS Step Functions is a low-code visual workflow service. It is perfect for GenAI for three reasons:

1. Automatic Retries: If Bedrock is throttled, Step Functions handles the retry logic automatically.
2. Visual Debugging: You can see exactly which stage of the AI process failed.
3. Wait States: If an AI job takes 5 minutes (Async pattern), the Step Function can "wait" for a callback without incurring compute costs.

---

3. Workflow Architecture: The AI Pipeline

graph TD
    A[Start: User Request] --> B[Lambda: Pre-process & Format]
    B --> C{Bedrock: Extract Intent}
    C -->|Complex| D[AWS Step Function: Multi-stage Analysis]
    C -->|Simple| E[Lambda: Direct Response]
    
    subgraph Step_Function_Detail
    D1[Task 1: Retrieve Docs] --> D2[Task 2: Summarize]
    D2 --> D3[Task 3: Reality Check]
    end
    
    D3 --> F[End: Send Result]
    E --> F

---

4. Error Handling and "Human-in-the-Loop"

One of the most powerful features of Step Functions in GenAI is the Task Token pattern.

• Scenario: The AI generates a response, but it has low confidence.
• Step Function Logic: The workflow pauses, sends an email to a human via Amazon SNS, and waits for the human to click "Approve" or "Reject" before finishing the workflow.

This is a core requirement for Domain 2 (Implementation) and Domain 3 (Safety).

---

5. Professional Implementation: The Map State

What if you have a 1,000-page document? You can use the Inline Map State in Step Functions to:

1. Break the document into 50 chunks.
2. Call Bedrock 50 times in Parallel.
3. Aggregate the results into a single "Reductive Summary."

The Result: You finish in 30 seconds instead of 15 minutes.

---

6. Code Fragment: Defining a Step Function Task for Bedrock

While you can write the JSON (ASL), the exam focuses on the logic.

{
  "StartAt": "CallBedrock",
  "States": {
    "CallBedrock": {
      "Type": "Task",
      "Resource": "arn:aws:states:::bedrock:invokeModel",
      "Parameters": {
        "ModelId": "anthropic.claude-3-sonnet-20240229-v1:0",
        "Body": {
            "anthropic_version": "bedrock-2023-05-31",
            "messages": [{"role": "user", "content": "Analyze input data."}]
        }
      },
      "Retry": [
        {
          "ErrorEquals": ["Bedrock.ThrottlingException"],
          "IntervalSeconds": 2,
          "MaxAttempts": 3,
          "BackoffRate": 2.0
        }
      ],
      "End": true
    }
  }
}

---

Knowledge Check: Test Your Orchestration Knowledge

---

Summary

Don't build "Mega-Prompts." Build "Micro-Steps." By using Step Functions, you turn a brittle AI call into a resilient business process. In the next lesson, we look at API Design and Integration Best Practices for these complex workflows.

---

Next Lesson: The Interface of AI: API Design and Integration Best Practices