Week 2 - Workflows and Agents

Understanding LLM Architecture Patterns

Lesson Overview

Segment Duration
Lecture: Agents and Workflows 5-10 minutes
Activity: Workflow vs. Agents 20 minutes
Exploration: Workflows 20+ minutes

Learning Objectives: By the end of this lesson, students will be able to:

  • Explain common features of LLM providers
  • Explain the difference between workflows and agents
  • Identify when to use one or the other
  • Plan workflows and agents conceptually
  • Translate workflows into code

Colab Notebook for Today:

Week 2 - Workflows and Agents
(It is recommended to download a copy of the notebook to your own google colab)

Lecture (5-10 min): Agents and Workflows

  • Show visually and explain the difference between a model, chat, agent, and workflow

(Optionally:)

  • Most common features of LLM providers
    • system instructions
    • temperature
    • function calling
    • structured output
    • multimodality
    • image, audio, and video generation

To get an idea of the breadth of what is offered, look through the left side of these pages

Activity (20 min): Workflow vs. Agents

Reading (5 min)

Spend 5 minutes, pick an article or two from the links below and start glancing through it to introduce workflows and agents. We will reconvene to talk through the differences together.

Guiding Questions:

  • What are the strengths and weaknesses of workflows vs agents?
  • In what situation is one better than the other?
  • When would a hybrid approach work best?

Prompt vs. Workflow vs. Agent: Great pro/con comparison, entertaining voice

Agent Patterns: Great depth into diagramming and specific use cases, Gemini implementation

LangChain on workflows and agents: Specific LangGraph implementation, brief

Discussion (2 min): Workflow vs Agents

Talk with the group about what you learned for the guiding questions.

Activity (10 min)

After looking through the articles, identify how you would diagram the following use cases:

  • Self assesses its own responses and redelivers.

  • Changes tone based off of user’s emotional state.

  • Database retrieval.

  • Extract text from a document, convert it to markdown, then summarize it. Return both the markdown and the summary.

  • Tiered LLM usage: Routing simple queries to faster, cheaper models (like Llama 3.1 8B) and complex or unusual questions to more capable models (like Gemini 3 Pro).

  • Generating multiple perspectives: Asking multiple LLMs the same question with different persona prompts and aggregating their responses.

  • Writing and refinement: Generating a draft, reflecting on its clarity and tone, and then revising it.

  • Booking appointments using a calendar API.

  • Controlling smart home devices.

STOP

Help someone that is behind catch up, either with the current lesson, or a past one.

The exploration activity invites branching into one of two groups: Visual or Code. If you’d like to work together (or find help when it is needed), find someone branching in the same direction.

When the people around you are ready, start exploring!

Exploration: Workflows (20+ min)

  • Option 1: Workflow with Graphical Interface (n8n, LangFlow).

  • Option 2: Workflow with Code (LangGraph).

  • Option 3: Skip workflows and just dive into coding agents (next lesson).

You can find the first two options below

Option 1: Visual Workflow Editors (n8n, LangFlow)

n8n and LangFlow are both free and popular. To choose between the two:

  • n8n is great for easy integration with products (Slack, Gmail, Salesforce, Notion) and for moving data between them.

  • LangFlow is similar but built on LangChain, letting you modify the code it creates and have finer control over the application.

n8n

To get started: install it, open it, and play around.

If you have Node.js installed already, this will get you up and running quick

# The % means it is a terminal command
%npm install n8n -g

# Once installed, run it
%n8n start

If not, you have two options:

  • Use the cloud version (you will sign up for a free account by clicking “Get Started for Free”)

  • Install Node.js then run the above code (Node.js is Javascript outside a web browser)

Either way, you can continue with these resources when you have explored it a little.

LangFlow

Warning: You may have to wait a while for the program to download if you go this route

  • Step 1: Install it as an app from their website

  • Step 2: Go through a Quickstart tutorial

There is an alternative installation route using the command line that you can find here

Option 2: LangGraph (Python)

Try running the code from the earlier LangChain article or exploring their docs.

Choose Your Platform

You can follow along using either Google Colab (recommended for beginners - no setup required!) or local development (requires Python environment setup). The code examples below include instructions for both.

Setup for Google Colab

First, set up your API key using Colab’s secrets (same as Week 1):

  1. Click the 🔑 key icon in the left sidebar (“Secrets”)
  2. Add a secret named GOOGLE_API_KEY with your Gemini API key
  3. Toggle on “Notebook access”
# Set up API key from Colab secrets
import os
from google.colab import userdata

os.environ['GOOGLE_API_KEY'] = userdata.get('GOOGLE_API_KEY')

Install required packages:

!pip install -q -U langchain langchain-google-genai langgraph

For Local Development:

# Load environment variables from .env file (local development only)
from dotenv import load_dotenv
load_dotenv()

Just replace their first two code chunks with this more relevant code:

from langchain.chat_models import init_chat_model

# Instantiate model
llm = init_chat_model(
    model="gemini-2.5-flash", 
    model_provider="google_genai"
)