Building your first AI agent with Make

Introduction

Ready to add intelligent automation to your workflows - without writing any code? Make’s AI Agents empower you to build goal-driven automations that think and choose actions using the tools you define. Here’s how you can create your first AI agent with clarity and ease.

1. What Is an AI Agent in Make?

Make AI Agents are modular, flexible automations driven by a large language model (LLM). Instead of running a fixed sequence, the agent evaluates your prompt, picks from a set of scenarios (tools) to act, and returns a structured response. Each run is stateless - every task is handled independently - which means crisp instructions and tool descriptions are key to consistent behavior.

2. Step-by-Step: Building Your First Agent

Step 1: Create Your AI Agent

• Create your account and navigate to AI Agents in the Make interface and click Create agent.

• Name your agent (e.g., Inventory Assistant) and provide a direct system prompt outlining its purpose and constraints - like checking stock or placing orders.

• Choose your AI provider (like OpenAI, Anthropic, etc.), connect via API key, and save your agent.

Step 2: Build and Attach Tools (Scenarios)

Design one scenario per tool. Examples:

• List Inventory: Pulls stock data from a source (e.g., Google Sheets), accepts a product name, and returns structured stock details.

• Order Stock: Generates restock requests or new order entries based on inputs like item and quantity. Ends with a return output.

• Give each scenario a clear name and purpose, and importantly, ensure it ends with a Return output module for structured results.

• Move to your agent’s settings and add these scenarios as tools - making them available for use.

Step 3: Connect an Input Channel

Create a scenario to serve as your agent’s front-end:

• For instance, use Slack → Watch messages to trigger on new messages, filter out bot messages, then use the Run an agent module.

• Map the incoming message to the agent prompt, and use Slack → Create a message to relay the response back—threaded, of course.

• Other possible triggers include Tally, Telegram, email, Teams, etc. Make supports many integrations.

Step 4: Test and Refine

• Try direct prompts like “Do we have blue hoodies?” or “Order 10 bags of coffee.”

• Introduce edge-case phrasing such as “Running low on espresso beans?” and ensure the agent handles them cleanly or asks for clarification.

• Check the scenario run history: which tool fired, what inputs/outputs were exchanged - this helps identify issues.

3. Essential Best Practices

• Keep agent scope focused: Begin with a single, clearly defined use case before expanding.

• Write concise tool names and descriptions - agents rely on these to pick the correct tool.

• Define inputs and outputs clearly, including types and requirements, so tools behave predictably.

• Debug via scenario logs: adjust prompt wording or tool metadata as needed

• Limit tools available at first: too many choices can confuse the agent.

• Use context files wisely: upload internal reference docs (e.g., style guides) to improve decision-making, but avoid sensitive data exposure.

• Manage cost and token usage:

• Control output tokens, agent steps, and history length.

• Start with a fast, affordable LLM model—e.g., OpenAI’s GPT-4 Mini.

4. Watch and Learn: Make AI Agent Demo

Here’s a video demonstrating Make’s AI Agent in action.

View embed

Final Thoughts

Building your first AI agent with Make is a smooth, no-code process:

• Define the agent’s role and language model.

• Build clear, single-purpose tool scenarios.

• Wire up an easy input method.

• Test thoroughly and refine.

Keep scope narrow, descriptions specific, and token costs in check.