Exercise: Module 1, Lesson 3 - Designing an Agentic Workflow

Objective: To practice the principles of agentic design by conceptually mapping out the workflow for a new AI agent, without writing any code. This exercise focuses on the "whiteboard" or planning stage of agent development.

Your Task

Imagine you are tasked with designing a simple "Research Assistant" agent. The agent's goal is to take a user's question, research it online, and provide a summarized answer with sources.

Your task is to design the agent's workflow by defining its components, similar to how you would structure a graph in a framework like LangGraph.

Deliverable

Create a Markdown file that describes your agent design. Your design must include the following sections:

1. Agent's Goal:

• A single, clear sentence describing the agent's primary objective.

2. Tools Required:

• A list of the external "tools" the agent will need to perform its job. For each tool, give it a name (e.g., web_search) and a brief description of what it does (e.g., "Takes a search query and returns a list of URLs and snippets.").

3. Nodes in the Graph:

• A list of the key "nodes" or steps in your agent's workflow. Each node represents a function or a point of reasoning. You should have at least three nodes (e.g., planner, researcher, summarizer).

4. Edges and Workflow Logic:

• Describe the flow of logic (the "edges") that connect your nodes. Explain the step-by-step process the agent will follow from the moment it receives a user's question to when it delivers the final answer. How does the state change as it moves from node to node? When does it decide to use a tool?

Example Submission Snippet:

1. Agent's Goal:

To answer a user's question by searching the web, synthesizing the findings, and providing a concise summary with source URLs.

2. Tools Required:

• web_search: A tool that takes a string search query and returns a list of relevant URLs.
• web_scraper: A tool that takes a URL and returns the full text content of the page.

3. Nodes in the Graph:

• planner: Takes the user's initial question and creates a step-by-step research plan.
• search_and_scrape: Executes the research plan by calling the web_search and web_scraper tools to gather information.
• synthesizer: Takes the gathered information and the original question, and generates a final, summarized answer.

4. Edges and Workflow Logic:

1. The workflow begins at the planner node. It receives the user's question (e.g., "What are the benefits of a 4-day work week?") and generates a research plan (e.g., "1. Search for 'benefits of 4-day work week'. 2. Scrape the top 3 URLs. 3. Synthesize the findings.").
2. The output of the planner (the plan and the list of URLs) is passed to the search_and_scrape node. This node executes the plan, calling the web_search tool, and then passing the resulting URLs to the web_scraper tool to get the full text of the articles.
3. The scraped text is then passed to the synthesizer node. This node takes the raw text and the original user question and calls the LLM with a prompt like: "Based on the following articles, answer the question: 'What are the benefits of a 4-day work week?' Provide a summary and list the source URLs."
4. The output of the synthesizer is the final answer, which is then returned to the user.