The Definitive Guide to Building Autonomous AI Agents with Make (Video Tutorial)

The Definitive Guide to Building Autonomous AI Agents with Make (Video Tutorial)

Workflow automation has taken a significant leap forward. The introduction of AI agents within the Make platform provides a new layer of intelligent, autonomous capability to your automated processes. These are not simple, linear automations; they are dynamic systems that can reason, use tools, and manage complex tasks based on natural language instructions. This guide offers a comprehensive walkthrough of Make's AI Agents, from foundational concepts to advanced, real-world applications. You will learn the complete process of creating, configuring, and deploying these agents to handle tasks that were previously impossible without direct human intervention.

Start Automating for Free with Make

Start Automating for Free with Make

Understanding the Core Components of Make AI Agents

To build and deploy an effective AI agent in Make, you must understand its three fundamental components. Each part plays a distinct role in the agent's operation, and all three must be configured correctly for the system to function. These components are the agent itself, its tools, and a trigger mechanism.

1. The Agent: The Brain of the Operation

The agent is the central intelligence of your system. It is where you define the agent's purpose, behavior, and core instructions. This is created in a dedicated “AI Agents” section within your Make organization, separate from individual scenarios. This global availability means a single, well-defined agent can be used across numerous different automation workflows.

When creating an agent, you configure several key elements:

• Connection: You must connect the agent to a large language model (LLM) provider. Make supports connections to major providers like OpenAI, Google (for Gemini models), and Anthropic. You will need an API key for your chosen provider to establish this connection.
• Agent Name: A clear, descriptive name helps you identify the agent's function within your organization, such as “Lead Research Specialist” or “Customer Support Triage.”
• Model Selection: Based on your chosen provider, you will select a specific model. For instance, with OpenAI, you might choose between GPT-4, GPT-4o, or GPT-4 Turbo. The model choice can affect the agent's speed, cost, and reasoning capabilities.
• System Prompt: This is the most critical part of the agent's configuration. The system prompt is a detailed set of instructions that defines the agent's entire persona and operational logic. You use natural language to describe its role, primary functions, constraints, and the step-by-step processes it should follow to complete its tasks.

2. The Tools: The Agent's Skills and Actions

An agent's intelligence is only useful if it can interact with the outside world and perform actions. In Make, this is accomplished through “Tools.” A tool is simply a standard Make scenario that has been built to perform a specific, atomic task. For an agent to be able to use a scenario as a tool, the scenario must meet three specific criteria:

1. On-Demand Trigger: The scenario cannot be scheduled to run at regular intervals. It must be set to run “On-demand,” meaning it only executes when called by another process, in this case, the AI agent.
2. Defined Scenario Inputs: The scenario must be configured to accept data from the agent. These are the parameters the agent will provide when it decides to use the tool. For a tool that creates a task in Notion, the inputs might be “Task Title” and “Task Description.”
3. Defined Scenario Outputs: The scenario must be set up to return a result to the agent after it finishes running. This feedback loop is what allows the agent to understand the outcome of its action. For the Notion tool, the output could be the “URL of the new task,” which the agent can then use in a subsequent step.

3. The Trigger: Bringing the Agent to Life

The agent and its tools are dormant until they are activated. This activation happens within a separate Make scenario that serves as the trigger. This “trigger scenario” starts with a module that watches for a specific event, such as a new message in a Slack channel, a new row in a Google Sheet, or a new submission from a web form like Tally.

Following the trigger module, you add the “Run an agent” module from the Make AI Agents application. This module is the bridge between your trigger event and your configured agent. Here, you select which of your globally defined agents you want to run and pass it the necessary information from the trigger, such as the text from a Slack message or the content from a form submission. This is the final piece that connects the components, creating a fully functional, event-driven AI agent system.

A Step-by-Step Guide to Creating Your First AI Agent

Building an agent involves a structured process that moves from defining its identity to equipping it with the necessary skills and finally activating it within a workflow. Following these steps will ensure all components work together correctly.

H3: Configuring the Agent and System Prompt

Your first step is to navigate to the “AI Agents” section in the main Make menu. Here, you will see a list of all agents within your organization.

1. Create a New Agent: Start by clicking “Create agent.”
2. Establish a Connection: From the “Connection” dropdown, select your preferred LLM provider. If you have not already, you will be prompted to add a new connection using your provider's API key.
3. Select a Model: Choose the specific language model you wish the agent to use. For tasks requiring complex reasoning, a more advanced model like Claude 3 Opus or GPT-4o may be suitable. For simpler, faster tasks, a model like Claude 3 Sonnet or Haiku could be more efficient.
4. Write the System Prompt: This is where you bring the agent to life. Be explicit and detailed. A good practice is to structure the prompt with clear sections.
   • Agent Role and Purpose: Start with a high-level description. For example: “You are an expert executive assistant tasked with helping manage specific business operations via Slack.”
   • Primary Functions: List the main capabilities of the agent as a numbered or bulleted list. This helps the agent categorize its skills. Example: “1. Enroll new students in courses. 2. Log content ideas. 3. Create tasks in Notion.”
   • Task-Specific Instructions: For each function listed, provide a detailed breakdown of the process and the information required. For “Student Enrollment,” you might write: “Required information: Student's email address, First name, Last name, and Course selection. If any of this information is missing, politely ask the user for the missing details before proceeding.”
   • General Instructions: Add rules about the agent's behavior, such as tone of voice, error handling, and how it should confirm actions with the user.

After defining the system prompt, you can save the agent. It is now a reusable “brain” that you can call from any scenario.

Create Your Free Make Account

H3: Building Scenarios as Agent Tools

With your agent defined, the next step is to build the scenarios that will function as its tools. Each tool should be a separate, self-contained scenario designed for one specific purpose.

1. Create a New Scenario: For each tool, start a new scenario in the Make scenario editor.
2. Set the Trigger to ‘On-demand': The first module in your scenario should not be a standard trigger. Instead, right-click and select “Set up a trigger,” then choose “On-demand.” This makes the scenario callable.
3. Define Inputs: At the bottom of the scenario editor, click on the “Scenario inputs and outputs” icon. In the “Scenario Inputs” tab, define the data fields the agent needs to provide. For a tool to “Log a content idea,” you might add two text fields: idea_title and idea_description.
4. Build the Action Logic: Add the necessary modules to perform the tool's function. For the content idea logger, this would be an “Airtable: Create a Record” module. Map the scenario inputs (idea_title, idea_description) to the corresponding fields in your Airtable base.
5. Define Outputs and Return the Result: The final module must return data to the agent. Add the “Scenarios: Return output” module. In the “Scenario inputs and outputs” settings, go to the “Scenario Outputs” tab and define the data you want to send back. For the content idea tool, you could create a field called logged_idea_link. Then, in the “Return output” module, map the URL of the newly created Airtable record to this output field.
6. Add a Description: When you first add this tool to an agent, Make will ask you to provide a description. This is very important. The agent uses this short description to understand what the tool does and when to use it. Be clear and concise. For example: “Use this tool to log a new content idea to the content machine. The tool requires a title and a description for the idea.”

Repeat this process for every skill you want your agent to have. Building modular, single-purpose tools makes your system easier to manage and debug.

H3: Integrating the Agent into a Workflow

The final step is to create the trigger scenario that activates the agent. This scenario links an external event to your agent's decision-making process.

1. Choose a Trigger: Start a new scenario with the appropriate trigger module. For a Slack assistant, this would be “Slack: Watch New Events.” Configure it to watch for new messages in a specific channel.
2. Add the ‘Run an agent' Module: Search for the “Make AI Agents” app and add the “Run an agent” action module.
3. Configure the Module:
   • Agent: Select the agent you created (e.g., “Jan Slack Assistant”) from the dropdown list.
   • Agent-level tools: The tools you configured directly on the agent's main settings page will appear here.
   • Additional tools: You can add extra tools that are specific only to this scenario's execution. This allows for customization without altering the base agent.
   • Thread ID: This field is the key to providing your agent with conversational memory. You need to map a unique identifier for the conversation here. For Slack, the best practice is to use the “Thread Timestamp” if it exists, and if not, use the main message “Timestamp.” This groups all replies within a single thread into one continuous conversation for the agent.
   • Messages: This is where you pass the user's input to the agent. For a Slack trigger, you would map the “Text” element from the “Watch New Events” module here.
   • Additional system instructions: You can provide extra instructions that are only applied for this specific run, which are added on top of the agent's main system prompt.
4. Handle the Agent's Response: After the “Run an agent” module, add modules to act on the agent's output. The agent's final reply will be available as the response element. For a Slack assistant, you would add a “Slack: Create a Message” module, set it to reply in the same thread by mapping the “Thread Timestamp,” and map the response from the agent module into the text content of the Slack message.

Practical Use Cases for Make AI Agents

The true power of AI agents is revealed through their application to real business problems. These examples demonstrate how agents can take over complex, multi-step processes.

H3: The Autonomous Lead Research and Enrichment Agent

Manually researching new leads is a time-consuming task for any sales team. An AI agent can automate this entire process, delivering fully enriched contact profiles without any manual work.

The Workflow:

1. Trigger: The process begins when a new lead signs up via a Tally form. This triggers the Make scenario.
2. Initial Save: The first step saves the basic lead information (name, company, email) to HubSpot, creating a new contact record.
3. Agent Activation: The “Run an agent” module is triggered, activating the “Lead Research Agent.” The HubSpot Contact ID is passed to the agent.

The Lead Research Agent's Configuration:

• System Prompt: The prompt instructs the agent that its objective is lead cultivation and enhancement. It is told it will receive a person's name, company name, and email and that its goal is to find additional information to update their contact details. The workflow is broken down into steps: find a personal LinkedIn profile, find a company LinkedIn profile, extract data from those profiles, and update the CRM.
• Tools:
  • Google Search: This tool takes a search query as input (e.g., “Jan Meinecke 9x LinkedIn Profile”) and returns the top search result URLs.
  • Get LinkedIn Profile Details: This tool uses a service like Apify to take a LinkedIn profile URL as an input and extracts detailed information like job history, education, skills, and company size.
  • Extract content from a website: This tool can scrape a company's website to find information about their services or mission.
  • Update HubSpot Contact: This tool takes a HubSpot Contact ID and a collection of new data points (job title, company size, country) and updates the corresponding contact record.

The agent intelligently uses these tools in sequence. It first uses Google Search to find the LinkedIn URL. Then, it passes that URL to the “Get LinkedIn Profile Details” tool. Finally, it takes all the extracted information and uses the “Update HubSpot Contact” tool to enrich the record. The sales team is presented with a complete profile, ready for outreach.

Start Automating for Free with Make

H3: The Centralized Brand Guideline Agent for Content Generation

Maintaining a consistent brand voice across all marketing content can be a challenge, especially when using AI to generate drafts. A “Brand Guideline Agent” can act as a centralized editor, ensuring all AI-generated content adheres to your company's style.

The Concept:
This agent is unique because it may not have any tools at all. Its power lies entirely in its detailed system prompt.

• System Prompt: The prompt is loaded with your company's complete brand style guide.
  • Overall Voice: Define the desired tone (e.g., “Friendly and approachable, like talking to a knowledgeable friend. Maintain a guiding tone, offering clear advice rather than sounding like an expert or guru.”).
  • Tone & Personality: Detail the nuances (e.g., “Encouraging and supportive. Balance professional credibility with a warm, personable touch.”).
  • Formatting Guidelines: Specify rules for headlines, subheadings, bullet points, and calls-to-action.
  • Workflow: Instruct the agent on how to review, edit, or generate copy. “Review generated copy: Ensure adherence to the 9x Style Guide. Edit/Adjust sections as needed with explicit reasons.”

The Application:
This Brand Guideline Agent can be used as a layer in other content creation scenarios.

1. Trigger: A scenario to “Generate Blog Articles” is triggered when a new blog topic is added to a Google Sheet.
2. Primary Agent: A “Blog Article Agent” is run. Its job is to research and write the article.
3. Layered Instructions: In the “Run an agent” module for the Blog Article Agent, the “Additional system instructions” are used to tell it to consult another agent. The instruction would be: “Your role is to write a blog post on the provided topic. Before finalizing the content, you must ensure it aligns perfectly with the 9x Brand Style Guide. Use your tools to research the topic, then craft the article following the brand guidelines.”

In this setup, the Blog Article Agent first uses its own tools (Google Search, etc.) to do research. Then, when it synthesizes the information into a blog post, its core instructions are influenced by the comprehensive style guide defined in the Brand Guideline Agent. This creates a powerful, two-tiered system where one agent provides the writing skill and another provides the brand oversight, resulting in high-quality, on-brand content, every time. The final step in the scenario is to take the agent's output and create a new Google Doc, delivering a ready-to-review draft.

Start Automating for Free with Make

More Articles for you:

• ..ReactOS the open-source alternative to Windows has been updated.
• ..AIPodcasts Review: Is This AI the Podcast Game Changer Marketing Pros Have Been Waiting For?
• ..Vidko Review: Capitalize on the Growing Need for AI Video Creation and Establish a Profitable Venture with This Must-Have Video Generation Tool—Use It Yourself or Resell for Maximum Profits
• ..8 Common Email Marketing Pitfalls and Their Solutions
• ..Ways to optimize your use of Gemini Deep Research

FTC Affiliate Disclaimer: This article contains affiliate links. I may earn a commission from qualifying purchases made through these links.