This 20+ AI Agent Team Automates ALL Your Work (GPT-01) (Relevance AI)

Key Concepts

• AI Agent Team: A group of specialized AI agents working together to automate tasks.
• Director Agent: The main agent that receives user requests and delegates tasks to other agents.
• Manager Agents: Agents responsible for orchestrating and evaluating the work of sub-agents.
• Sub-Agents: Specialized agents with specific tools and capabilities for performing individual tasks.
• Tools: Software integrations or functionalities that allow agents to interact with external systems.
• Relevance AI: An AI agent and tool building platform used to create and manage the agent team.
• Make.com: A workflow automation tool used to integrate Relevance AI with other software.
• Agent Prompting: Crafting instructions for AI agents to guide their behavior and decision-making.
• Google Ser API: An API for accessing Google search results, including flights, hotels, and other data.
• Fine-tuning: Training AI models on specific datasets to improve their performance on particular tasks.

1. Overview of the 20 AI Agent Team

• The video demonstrates a system of 20 AI agents designed to manage and automate tasks across a user's tech stack.
• The system has access to communication channels (WhatsApp, LinkedIn, email, calendar, Slack, voice calling), project management tools (CRM, Notion, Google Docs, Google Drive), research capabilities, and content creation tools.
• The agent team can automate complex workflows involving multiple software platforms using simple English instructions.
• Examples include: finding flight options and sending them via WhatsApp, rescheduling meetings and updating calendars, researching leads and adding them to a CRM, and creating and publishing content on social media.
• The template for this system will be available in the creator's community.

2. Demo of the Agent Team in Action

• The user interacts with the agent team via voice messages on WhatsApp.
• Example 1: Retrieving Unread Messages and Calendar Events:
  • The user requests the agent to retrieve unread messages from all communication channels and list calendar events in a Google Doc.
  • The Director Agent delegates the task to the Communication Manager.
  • The Communication Manager instructs sub-agents for WhatsApp, LinkedIn, Slack, email, and calendar to retrieve the information.
  • The Project Manager creates a Google Doc with the information and sends it back to the Director Agent, who then sends it to the user.
• Example 2: Finding Flight Options and Sending to a Contact:
  • The user asks the agent to find three cheap flight options from Sao Paulo to Amsterdam for December 19th, put them in a Google Doc, and send them to their mom on WhatsApp, asking if the arrival times are suitable.
  • The agent uses the Google Ser API to find flight options.
  • The agent sends the flight options to the user's mom via WhatsApp and sends the Google Doc back to the user.
• Example 3: Researching AI Coding Agents and Creating Content:
  • The user instructs the agent to research the latest trends in AI coding agents and write a blog post and LinkedIn post.
  • The agent delegates the task to the Content Manager, who uses specialized agents for writing blog posts and LinkedIn posts.
  • The agent posts the blog post to the user's website and adds the LinkedIn post to the content calendar on Notion.
• Example 4: Lead Management:
  • The user asks the agent to research a LinkedIn lead, scrape their LinkedIn profile, add the enriched data to the CRM, and send a follow-up message.
  • The agent adds the lead to the CRM with enriched details and sends a message via Slack.
• The agent can also make phone calls to reschedule meetings and update the calendar with the outcome.

3. System Architecture and Components

• The system is built primarily in Relevance AI, with Make.com used for integrations not natively available in Relevance AI.
• The architecture consists of triggers, a Director Agent, Manager Agents, and Sub-Agents.
• Triggers: Initiate workflows, such as receiving a WhatsApp message or a scheduled event.
• Director Agent: The central point of contact, responsible for breaking down queries, planning tasks, delegating to manager agents, evaluating results, and communicating with the user.
• Manager Agents: Oversee specific areas like communication, project management, research, and content creation. They orchestrate and evaluate the work of sub-agents.
• Sub-Agents: Specialized agents with specific tools for interacting with software and performing tasks.
• The system uses 20 AI agents and over 50 tools.
• The design prioritizes reducing the responsibilities of individual agents to improve reliability.

4. Responsibilities of Each Agent Type

• Director Agent:
  • Breaking down user queries into sub-tasks.
  • Planning the workflow and delegating tasks to appropriate manager agents.
  • Evaluating the work done by manager agents and sub-agents.
  • Communicating with the user via WhatsApp.
• Manager Agents (e.g., Communication Manager):
  • Orchestrating and delegating tasks to sub-agents.
  • Evaluating the work of sub-agents and ensuring accuracy.
  • Communicating with the director agent.
• Sub-Agents (e.g., WhatsApp Agent):
  • Performing specific tasks using their assigned tools (e.g., sending WhatsApp messages, retrieving unread messages).

5. Tools and Integrations

• Communication Tools: WhatsApp, LinkedIn, email, calendar, Slack, voice calling.
• Project Management Tools: CRM (HubSpot), Notion, Google Docs, Google Drive.
• Research Tools: Google Search API, web scraping, LinkedIn scraping.
• Content Creation Tools: Webflow, LinkedIn, Act.
• Google Ser API: Used for finding flight options, hotel options, and other data.
• Make.com: Used to integrate Relevance AI with software that doesn't have native integrations.

6. Step-by-Step Breakdown of a Workflow (Retrieving Unread Messages)

1. Trigger: User sends a voice message to the Director Agent via WhatsApp.
2. Director Agent:
   • Receives the voice message (converted to text).
   • Identifies the need to use the Communication Manager.
   • Prompts the Communication Manager to retrieve unread messages from all communication channels.
3. Communication Manager:
   • Instructs the WhatsApp, LinkedIn, Slack, and email sub-agents to retrieve unread messages.
   • Instructs the calendar agent to retrieve calendar events.
4. Sub-Agents:
   • Retrieve the requested information from their respective channels.
5. Communication Manager:
   • Collects the information from the sub-agents.
   • Sends the collected information back to the Director Agent.
6. Director Agent:
   • Sends the information to the Project Manager.
7. Project Manager:
   • Instructs the Google Docs agent to create a Google Doc with the information.
8. Google Docs Agent:
   • Creates the Google Doc.
   • Sends the Google Doc link back to the Project Manager.
9. Project Manager:
   • Sends the Google Doc link back to the Director Agent.
10. Director Agent:
    • Sends a summary of the unread messages and the Google Doc link to the user via WhatsApp.

7. Agent Prompting and Configuration

• Agent Profile: Contains the agent's name and description, which is used by manager agents to understand the capabilities of sub-agents.
• Agent Instructions (Core Instructions): The main prompt that defines the agent's role, objective, and SOP.
• SOP (Standard Operating Procedure): A detailed breakdown of the agent's tasks.
• Examples: Provide the agent with examples of how to approach specific queries.
• Notes: Used to reinforce important rules and instructions.
• Tools: Define the software integrations and functionalities that the agent can use.
• Sub-Agents: List the sub-agents that the agent can delegate tasks to.
• Prompt for How to Use: Provides more context to the director agent on what the sub-agent can do.
• Template for Communication: Defines the prompt that the director agent should use when communicating with the manager agent.

8. Scheduling Messages and Automating Workflows

• Workaround using Make.com:
  1. Create a new scenario in Make.com.
  2. Use an HTTP request module to call the Relevance AI agent's API endpoint.
  3. Set the method to POST and add the necessary headers (Content-Type and Authorization).
  4. Set the body type to raw and the content type to JSON.
  5. Include the agent ID and conversation ID in the request content.
  6. Define the message to be sent to the agent in the content field.
  7. Schedule the scenario to run at regular intervals (e.g., every day at 8:00 AM).

9. WhatsApp Integration

• WhatsApp Business API: Requires setting up the WhatsApp Business API.
• Alternative using Relevance AI Premium Triggers: Connect a personal WhatsApp account to the agent and trigger it through text messages.
• Make.com Scenario:
  • Receives messages from the WhatsApp Business API.
  • Downloads media types (voice messages, images, documents).
  • Converts voice messages to text using speech-to-text.
  • Describes images using image-to-text.
  • Transcribes documents.
  • Sends the processed information to the Relevance AI agent.

10. Conclusion

• The video demonstrates a powerful AI agent team capable of automating complex workflows across various software platforms.
• The system's modular architecture, with specialized agents and tools, allows for flexibility and scalability.
• The use of Relevance AI and Make.com enables seamless integration and automation.
• The ability to schedule messages and automate workflows on a repetitive basis further enhances the system's efficiency.
• The creator emphasizes the importance of clear agent prompting, detailed SOPs, and providing context to the agents to ensure accurate and reliable performance.