How I’d Teach a 10-Year Old to Build AI Agents (n8n, No Code)

Key Concepts

• Large Language Models (LLMs): AI models like ChatGPT, Claude, Gemini, and DeepSeek trained on vast datasets to generate text-based outputs.
• Workflow Automations: Automated sequences of actions triggered by specific events, connecting various tools and applications.
• AI Agents: Autonomous entities that can think, make decisions, remember past interactions, and utilize tools to achieve specific goals.
• System Prompt: Instructions given to an AI agent to guide its behavior and decision-making process, similar to training an employee.
• Nodes: Building blocks in NADEN that represent actions or triggers in a workflow or AI agent.
• Triggers: Events that initiate a workflow or AI agent, such as receiving a chat message.
• Actions: Tasks performed by a workflow or AI agent after being triggered, such as sending an email or searching for a contact.
• Credentials: Securely stored authentication information (e.g., API keys) that allows NADEN to access and interact with external tools and services.

Large Language Models (LLMs)

• LLMs like ChatGPT, Claude, Gemini, and DeepSeek are trained on massive datasets.
• They receive a message as input and generate an output based on their training.
• Limitation: LLMs cannot directly interact with external tools or automate tasks without manual input.
• Example: Typing "Please reply to an email from Jonno confirming coffee tomorrow at 2 p.m or 3 p.m." into ChatGPT.

Workflow Automations

• Workflow automations automate processes by connecting different tools and applications.
• Example: Automatically receiving an email and generating a reply using AI.
• They can access tools like Gmail, ChatGPT, Google Sheets, ClickUp, Notion, and Slack.
• Limitation: Workflow automations are rigid and follow a strict "if this, then that" logic, lacking autonomy and decision-making capabilities.
• Analogy: Workflow automations are like following an IKEA manual step-by-step; if a step fails, the entire process halts.

AI Agents

• AI agents are presented as virtual employees that can think, act autonomously, and remember past interactions.
• They receive instructions through a system prompt, which is analogous to training an employee.
• AI agents have a "brain" (LLM) and "memory" to make decisions and recall past messages.
• They can access and utilize various tools to perform tasks.
• Example: An AI agent can be instructed to "send off an email," and it will autonomously handle the task.
• AI agents can connect to tools like Gmail, Google Sheets, Google Calendar, CRMs, ClickUp, Facebook, Twilio, Google Drive, Slack, and Notion.
• Key Difference from Workflow Automations: AI agents can think autonomously and make rational decisions to achieve goals, whereas workflow automations are rigid and predefined.
• Analogy: Unlike workflow automations, AI agents can troubleshoot and adapt when encountering problems, similar to how a human employee would handle unexpected issues.

Building an AI Agent in NADEN

• NADEN is presented as a free platform for building AI agents.
• Step 1: Create a NADEN Account: Sign up for a free account on NADEN.io.
• Step 2: Create a New Workflow: Click the "+" button, go to "Workflows," and select "Personal."
• Step 3: Add a Trigger: Use the "On Chat Message" trigger to initiate the AI agent when a message is received.
• Step 4: Add an AI Agent Node: Add an "AI Agent" node from the "AI" category.
• Step 5: Configure the AI Agent:
  • Chat Model (Brain): Choose a large language model (e.g., OpenAI). If you don't have credits, create a new credential using an API key from platform.openai.com.
  • Memory: Add "Simple Memory" to enable the AI agent to remember past messages.
  • Tools: Add tools like "Email" (Gmail) and "Google Contacts" to enable the AI agent to perform tasks.
• Step 6: Configure Tools:
  • Gmail: Connect your Gmail account by creating a new credential and granting permissions. Choose the "Send Email" operation and configure the "To," "Subject," and "Message" fields.
  • Google Contacts: Connect your Google Contacts account. Use the "Get Many" operation to search for contacts by name. Configure the query to extract the name from the incoming message.
• Step 7: Create a System Prompt:
  • Use ChatGPT to generate a system prompt that instructs the AI agent on how to use the available tools.
  • Example Prompt: "You are an AI agent integrated into an editing workflow. You have access to two tools: send an email via Gmail and search emails via Google contacts. Before sending any email, always search for and confirm the recipient's email address using the Google contact search tool. Only proceed with sending the email once a valid address has been found."
• Step 8: Test the AI Agent: Send a message to the AI agent and observe its behavior. Check the logs to see how it processes the message and uses the tools.

Example Scenario: Sending an Email to Jonno

1. Input: "Please send Jonno an email confirming coffee tomorrow at 2 p.m."
2. AI Agent Processing:
   • Receives the message.
   • Checks for past historical messages.
   • Uses its "brain" (LLM) to understand the task.
   • Uses Google Contacts to find Jonno's email address.
   • Uses Gmail to send the email.
   • Generates a response: "I've sent the email confirming coffee tomorrow at 2 p.m. Is there anything else you want me to do?"
   • Updates its memory with the conversation context.

Conclusion

The video explains the evolution from LLMs to workflow automations and finally to AI agents, highlighting the increasing levels of autonomy and decision-making capabilities. AI agents, built using platforms like NADEN, can act as virtual employees, leveraging tools and memory to perform complex tasks. The key to building effective AI agents lies in providing clear instructions through system prompts and connecting them to the right tools. The example of building an AI agent that can find email addresses and send emails demonstrates the practical application of these concepts.