Published: June 2026 | Reading Time: 13 minutes | Day 6 of the AI Agents for Business Series
TL;DR: You don't need to be a developer to build a working AI agent. In this post, I show you exactly how to build your first AI agent using [Make.com](https://www.make.com) — step by step, screenshot by screenshot — so you can have a real automation running for your business by the end of this guide.
Okay, I have to be completely honest with you.
When I first heard the phrase "build an AI agent," I pictured myself writing thousands of lines of Python, setting up servers, and crying into a cup of cold coffee at 2am.
Turns out? It took me about 45 minutes. No code. No server. Just Make.com, a free OpenAI API key, and a workflow I now use every single day in my business.
In this guide — Day 6 of our AI Agents for Business series — I'm going to walk you through exactly how I did it. We're building a real AI agent that monitors your email inbox, classifies incoming messages, and automatically drafts personalised replies based on the type of enquiry.
If you haven't read the earlier posts in this series, I'd recommend starting with [what AI agents actually are](https://datalabbooks.com/what-are-ai-agents-the-complete-non-technical-guide-for-business-owners), then coming back here. But if you're ready to build right now — let's go.
Before we build, let me quickly explain why I chose Make.com over every other tool for this tutorial.
I've tested [n8n](https://n8n.io), [Voiceflow](https://www.voiceflow.com), [AgentGPT](https://agentgpt.reworkd.ai), and [Relevance AI](https://relevanceai.com). They're all great tools for different reasons (and I covered each one in [my tool comparison post](https://datalabbooks.com/best-ai-agent-tools-for-small-business-in-2026-tested-ranked)). But for someone who has never built an AI agent before, [Make.com](https://www.make.com) wins every time for three reasons:
It's visual. You build your agent by dragging and dropping modules on a canvas. You can see every step, every connection, every decision branch — all at once. There's no guessing what's happening behind the scenes.
It connects to everything. Make.com has 1,800+ app integrations out of the box. Gmail, Slack, Notion, Google Sheets, HubSpot, Airtable, WhatsApp, Stripe — you name it. Your agent can talk to virtually any tool your business already uses.
It has a free plan. You can build and test this entire agent without spending a penny. The free tier gives you 1,000 operations per month — more than enough to start.
What We're Building Today
Here's the exact agent we're going to build together:
The Email Triage Agent
Trigger: A new email arrives in your Gmail inbox
2. Step 1 — Read & Extract: The agent reads the email subject and body
3. Step 2 — Classify: An AI module (powered by OpenAI GPT-4o) classifies the email into one of three categories: Sales Enquiry, Support Request, or Other
4. Step 3 — Draft Reply: Based on the classification, the AI writes a personalised draft reply in your tone
5. Step 4 — Save to Google Sheets: The email details and classification are logged automatically
6. Step 5 — Alert: A Slack notification tells you: "New Sales Enquiry draft ready for review"
That's a genuine, production-ready AI agent. It perceives (reads email), reasons (classifies + drafts), acts (logs + notifies), and can be set up to observe and improve. Sound familiar? That's the exact [Perceive → Reason → Act → Observe loop](https://datalabbooks.com/how-ai-agents-make-decisions-the-plain-english-breakdown-2026) I broke down in last week's post.
What You'll Need Before We Start
This is the full setup checklist. All free to start:
A free [Make.com account](https://www.make.com) (sign up takes 2 minutes)
2. A Gmail account connected to Make.com
3. A free [OpenAI API key](https://platform.openai.com) (you'll get $5 free credit when you sign up)
4. A Google Sheets spreadsheet (we'll create it from scratch)
5. A Slack workspace (optional, but highly recommended)
That's it. No credit card. No coding environment. No engineering degree. Let's build.
Step-by-Step: Building Your First AI Agent in Make.com
Step 1: Create a New Scenario in Make.com
Log in to your Make.com account and click Create a new scenario from the dashboard. You'll land on the visual canvas — the blank grid where we'll build everything.
In Make.com, a scenario is your agent's workflow. Think of it as the blueprint that tells your agent what to watch for, what to think about, and what actions to take.
Click the big + button in the centre of the canvas to add your first module.
Step 2: Set the Trigger — "Watch Emails" in Gmail
Search for Gmail in the module search bar and select the Watch Emails trigger. This is the "Perceive" stage — your agent's eyes. It watches your inbox and fires every time a new email arrives.
Connect your Gmail account when prompted (Make.com uses secure OAuth — your password is never stored). Then configure:
Click OK to save the trigger. You'll see the Gmail module appear on your canvas as a circle. This is your agent waking up.
Step 3: Add the AI Brain — OpenAI "Create a Completion"
Click the + icon on the right side of your Gmail module to add the next step. Search for OpenAI and select Create a Completion. This is the "Reason" stage — the AI brain that reads the email and decides what it's about.
Connect your OpenAI API key. Then set up the module like this:
Model: gpt-4o (recommended for best classification accuracy)
- System Role: You are an expert email classifier for a small business. Your job is to read each email and classify it as exactly one of three categories: SALES_ENQUIRY, SUPPORT_REQUEST, or OTHER. Reply with only the category label, nothing else.
- User Message: Map this to the Gmail trigger — use {{subject}} and {{snippet}} from the email data
This system prompt is the secret sauce. It gives the AI a tight, specific instruction — exactly as I explained in my post on [how AI agents make decisions](https://datalabbooks.com/how-ai-agents-make-decisions-the-plain-english-breakdown-2026). The more specific your prompt, the more reliably your agent performs.
Step 4: Add a Router to Branch by Category
This is where Make.com's visual decision logic shines. Click + after the OpenAI module and select Router. A router splits your workflow into multiple branches based on conditions — this is how your agent "decides" what to do next.
Create three branches from the router:
Branch 1: Filter condition = OpenAI output contains "SALES_ENQUIRY"
- Branch 2: Filter condition = OpenAI output contains "SUPPORT_REQUEST"
- Branch 3: Filter condition = OpenAI output contains "OTHER"
Your canvas will now show three paths branching out from the router. Visually, you can see your agent's decision tree. This is one of the things I genuinely love about Make.com — the logic is completely transparent.
Step 5: Draft the AI Reply for Each Branch
On each branch, add another OpenAI Create a Completion module. This time, the agent writes the actual reply email. Here's the system prompt I use for the Sales Enquiry branch:
"You are a friendly, professional sales assistant for [Your Business Name]. A potential customer has sent us an email. Write a warm, personalised reply that acknowledges their enquiry, briefly explains what we offer, and invites them to book a 15-minute discovery call. Keep the tone conversational and never pushy. Max 150 words."
For the Support Request branch, adjust the prompt to acknowledge the issue and explain your support process. For Other, have it draft a polite generic acknowledgement.
Pro tip: Connect the Gmail {{from.email}} and {{subject}} fields into the user message so the AI can personalise the reply with the sender's name and reference the specific email they sent. It makes a huge difference.
Step 6: Log Everything to Google Sheets
After each branch's AI reply module, add a Google Sheets: Add a Row module. Create a spreadsheet with columns for: Date, Sender Email, Subject, Category, AI Draft Reply, Status.
Map each field from the earlier modules. Now every email your agent processes gets logged automatically. This is your agent's memory — a running record of everything it has seen and done. You can review it, spot patterns, and improve your prompts over time.
Step 7: Send a Slack Alert
For the Sales Enquiry branch specifically, add a Slack: Send a Message module after the Google Sheets log. Configure it to post to your #sales channel with something like:
"📧 New sales enquiry from {{from.email}} | Subject: {{subject}} | AI draft is ready in Google Sheets. Go review and send!"
This is the "Act" and "Observe" stage completing the loop. Your agent has processed the email, generated a response, logged it, and now told you about it. All within seconds of the email landing.
Step 8: Activate and Test Your Agent
Before turning your agent on, click Run once in the bottom-left corner of Make.com. This runs a single test without activating the full automation. Send a test email to yourself from a different account, then watch your scenario run in real time.
Each module on the canvas will show a green bubble with the number of records it processed. If something goes wrong, Make.com highlights exactly which module failed and why. No debugging nightmares.
Once your test passes, toggle the On/Off switch in the bottom left to On. Your agent is now live. It will check your inbox every 15 minutes (on the free plan) and process any new emails it finds.
Congratulations. You just shipped your first AI agent.
5 Ways to Level Up This Agent (Once It's Running)
Once your basic agent is live and you've watched it work for a few days, here's how to make it significantly more powerful:
1. Auto-Send Instead of Auto-Draft
Once you trust your agent's replies (I usually wait 2 weeks and review every draft first), swap the Google Sheets log step for a Gmail: Send an Email module. Your agent goes fully autonomous on routine enquiries — you only step in for the complex ones.
2. Add a CRM Integration
For Sales Enquiry emails, add a HubSpot: Create a Contact or Airtable: Create a Record module. Every new lead gets automatically added to your CRM with the email content, classification, and AI-drafted reply attached. Zero data entry.
3. Add a Sentiment Filter
Add a second OpenAI step that also classifies the email sentiment: Positive, Neutral, or Urgent/Frustrated. Use this to trigger a different reply template for upset customers — one that prioritises empathy over efficiency.
4. Build a Knowledge Base Connection
Connect your agent to a Notion database or Google Doc that contains your FAQs, pricing, and product information. Pass this context into your OpenAI prompt so the agent can give accurate, specific answers rather than generic ones. This transforms your agent from a classifier into a genuine knowledge assistant.
5. Add an Escalation Rule
Add a filter that checks whether the email contains keywords like "refund," "legal," "urgent," or "disappointed." If it does, skip the AI reply entirely and send yourself an immediate Slack DM to handle it personally. This is how you keep a human in the loop for high-stakes situations — exactly the kind of responsible agent design I talked about in my [AI agents vs automation post](https://datalabbooks.com/ai-agents-vs-automation-whats-the-real-difference-and-which-one-does-your-business-actually-need).
Common Mistakes to Avoid (I Made All of These)
Let me save you the frustration I went through when I built my first agent:
Mistake 1: A vague system prompt. If you just write "classify this email," the AI will return inconsistent responses — sometimes full sentences, sometimes just a word. Always be explicit about the exact format you want back.
Mistake 2: Not testing edge cases. What happens when your agent gets a spam email? A blank email? An email in a different language? Run 10 different test emails through before going live.
Mistake 3: Skipping the Google Sheets log. I know it feels like extra work, but this log is how you review your agent's performance, catch mistakes, and improve your prompts. Don't skip it.
Mistake 4: Setting the interval too short. Make.com's free plan runs scenarios on a schedule. Setting it to check every 1 minute will eat through your operations fast. 15 minutes is the sweet spot for most business email use cases.
Mistake 5: Trying to build everything at once. Start with the basic 4-module version first. Get it working. Then add one enhancement at a time. This is the same advice I gave in my post on [how AI agents work step by step](https://datalabbooks.com/how-ai-agents-work-a-complete-step-by-step-breakdown-2026-guide) — simple, focused agents outperform bloated, complex ones every time.
I get asked this a lot, so let me be clear about [Make.com's pricing](https://www.make.com/en/pricing):
| Plan | Monthly Cost | Operations/Month | Scenarios | Best For |
|---|---|---|---|---|
| Free | $0 | 1,000 | 2 | Getting started, testing |
| Core | $9/mo | 10,000 | Active scenarios | Small businesses |
| Pro | $16/mo | 10,000 | Unlimited | Growing teams |
| Teams | $29/mo | 10,000 | Unlimited | Multiple users |
For most small business owners, the Core plan at $9/month is more than enough to run 3-5 active agents handling hundreds of emails, leads, or tasks per month. That's less than the cost of one hour of a virtual assistant.
What to Build Next: 5 More AI Agent Ideas for Make.com
Once your email agent is running, here are five more Make.com agents I'd recommend building next — roughly in order of impact:
The Lead Qualification Agent — When someone fills out your website contact form, the agent scores their likelihood to buy, enriches their data from LinkedIn, and either adds them to a nurture sequence or flags them as a hot lead for immediate follow-up.
The Content Repurposing Agent — Every time you publish a blog post (like this one), the agent automatically creates 3 LinkedIn posts, 5 tweets, and an email newsletter summary. One input, multiple outputs, zero extra work.
The Invoice Chaser Agent — Monitors your accounting software for overdue invoices and sends polite, progressively firmer follow-up emails at day 3, day 7, and day 14 — all personalised to the client.
The Review Monitor Agent — Watches Google Reviews and Trustpilot for new reviews, classifies the sentiment, and drafts a personalised public reply for you to approve. Never miss a review again.
The Daily Business Briefing Agent — Every morning at 8am, sends you a Slack summary of yesterday's key metrics: emails received + classified, new leads, pending support tickets, and any urgent flags. Your personal AI business analyst.
Summary: What You Built Today
Let's recap what you now have running:
A fully functional AI Email Triage Agent built on [Make.com](https://www.make.com) that watches your Gmail inbox, uses GPT-4o to classify every incoming email, drafts a personalised reply for each category, logs everything to Google Sheets, and sends you a Slack alert for priority messages — all automatically, 24 hours a day, 7 days a week.
This is not a demo. This is not a toy. This is a real AI agent doing real work for your business right now.
And you built it without writing a single line of code.
More From This Series
If you've been following along from the beginning, here's the full series so far:
[What Are AI Agents? The Complete Non-Technical Guide for Business Owners](https://datalabbooks.com/what-are-ai-agents-the-complete-non-technical-guide-for-business-owners)
2. [Best AI Agent Tools for Small Business in 2026 (Tested & Ranked)](https://datalabbooks.com/best-ai-agent-tools-for-small-business-in-2026-tested-ranked)
3. [How AI Agents Work: A Complete Step-by-Step Breakdown](https://datalabbooks.com/how-ai-agents-work-a-complete-step-by-step-breakdown-2026-guide)
4. [AI Agents vs Automation: What's the Real Difference?](https://datalabbooks.com/ai-agents-vs-automation-whats-the-real-difference-and-which-one-does-your-business-actually-need)
5. [How AI Agents Make Decisions: The Plain-English Breakdown](https://datalabbooks.com/how-ai-agents-make-decisions-the-plain-english-breakdown-2026)
6. How to Build Your First AI Agent with Make.com ← You are here
Written by the DataLabBooks team. We publish plain-English guides for business owners using AI agents — no tech background needed. New post every day.
