AI agent: Goal oriented functionalities added to LLM (task completion) by leveraging tools. Makes multiple thoughts.

N8N: workflow automation tool to build workflows with multiple steps. Some of the steps can be an AI agent.

Prompting layers in Agentic Systems

• System layer: role, rules, how to complete task

  • Role prompting: explain who the agent is and not what it does
    • Format: You are a … Your role is …
  • Instructions:
    • Format: Follow the following steps to do XYZ: 1… 2…. 3…
  • Rules
    • Format: Always… Don’t…
  • Few shot prompting
    • Meaning: a shot is an example… one shot prompting is giving the AI one example.
    • Format: Use the following examples when replying to user…<examples><example>a single example</example></examples>
    • XML v Markdown: some chatmodels prefer xml and others prefer markdown…. i.e. <example> vs ## Example
  • Additional Context (Environmental)
    • Datetime ( {{ $now }} )
    • Name of business
    • Contact & phone number
  • Reducing halucinations
    • Read article of your model on reducing halucinations.
    • Example prompt: Don’t make things up, aske the user a clarifying question if you need additional information to complete your task. If you’re asked a question to which you don’t know the answer, say so.

• Input layer: specific task request

  • Can be a user message or another system.

• Action layer: defines tools and their inputs and outputs

  • Defines how AI can use the tools properly
  • Description
    • Format: Use this tool_name tool to fetch existing ….
    • Instructions part: To fetch … 1. set eventsAfter, eventBefore, otherVariablesTheToolNeeds… 2. The tool will return…. 3. Once you have… do XYZ

Core concepts

• Trigger: starts an automation
  • Manual
  • Scheduled
  • Applications
    • Webhook
    • Form submission
• Filtering: allow or block certain types of data
• Actions (Apps): allow to interact with other apps.

Best practices

• Mapping: planning to understand the tasks, apps, tools, actions,