Why Frameworks

• Key points are in the visual diagram above.

Raw while-loops work for a demo and collapse when you need checkpoints, handoffs, or human approval. Frameworks encode those production problems so you are not rewriting orchestration for every product.

Phase 4–7 taught you how an engine works. A framework is the car — you still need to know what a clutch does when it fails on the highway.

Scenario

A refund agent needs: classify → lookup order → maybe refund → human gate over $200. That is a graph (LangGraph), a typed tool loop (PydanticAI), or a crew with a manager. The framework is the wiring, not the policy.

Solution

In Agent Framework Landscape, apply Why Frameworks to this scenario: A refund agent needs: classify → lookup order → maybe refund → human gate over $200. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Why Frameworks (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Why Frameworks happens in the code.

1def agent_loop(user, tools, llm, max_steps=8):  # define a reusable function2    state = {"messages": [user], "step": 0}3    while state["step"] < max_steps:4        decision = llm.decide(state, tools)5        if decision.type == "final":6            return decision.text  # return the result7        state["messages"].append(tools.run(decision.tool, decision.args))8        state["step"] += 19    return "stopped: max steps"  # return the result10# Frameworks replace this loop with graphs, crews, or typed runners.

• Starting with LangGraph before writing a raw loop
• Treating MCP as a competitor to CrewAI
• Collecting frameworks instead of shipping one agent

Quick recap — the most important points from this module.