Module 10 Wrap-up: The Sovereign Decision Maker

You have graduated from "Building Pipes" (Chains) to "Building Minds" (Agents). You now understand that an Agent is just an LLM that has Descriptions of Tools in its system prompt and a Loop that allows it to try multiple times until it succeeds.

---

Hands-on Project: The Autonomous Market Research Agent

1. The Goal

Build an agent that can answer the following question: "What is the current stock price of Apple, and how does it compare to its price 5 years ago?"

2. The Implementation Plan

1. Tools: Use TavilySearchResults for the current price and a custom @tool (or Wikipedia) for historical data.
2. Model: Use ChatOpenAI (gpt-4o-mini).
3. Executor: Initialize an AgentExecutor with max_iterations=5.
4. Prompt: Use a standard "OpenAI Functions" prompt from the Hub.

---

Module 10 Summary

• Agents: Use LLMs to determine the sequence of steps.
• Reasoning: The ReAct cycle (Thought $\rightarrow$ Action $\rightarrow$ Observation).
• Tool Choice: The semantic matching of descriptions to user needs.
• Safety: Using max_iterations to prevent budget catastrophes.

Coming Up Next...

In Module 11, we refine the "Shape" of the agent's output. We will learn about Structured Output and Parsers, ensuring our agents return clean JSON that can be used by other software, not just human-readable text.

---

Module 10 Checklist

• I can describe the difference between a Chain and an Agent.
• I have initialized an AgentExecutor in a Python script.
• I have seen the agent call a tool in the "Verbose" logs.
• I have set a max_iterations limit to 10 or less.
• I understand that the agent doesn't "know" the tool's code, only its description.