How to Set Up an AI Agent for Your Business: n8n vs Claude Code

Hi, I'm Pavel, I help to automate business workflows. I'm starting a new series on how to set up AI agents for small businesses. And today we're going to set up AI agent to manage restaurants booking.

This is what the guest sees. She opens the restaurant Telegram, and asked for a table for six this Saturday at eight. And here's the workflow.

The agent gets this message, checks the restaurant booking table, sees that it's full at 8pm. and asks the guest alternative options 19:30 or 9. The guest decides 19:30.

Now the agent asks for approval from the manager. The manager approves and the booking is saved in the booking table and the confirmation message is sent to the guest. The whole thing took like 15 seconds.

How does it actually work? Every AI agent, no matter how impressive it looks, is built from the same four pieces. The first piece is the trigger.

In our case, that's a guest sending a message to the restaurant. The second piece is the brain, an AI model together with instructions written in plain language. Opening hours, the biggest group you accept online, how to handle allergies, the tone you want to speak, and so on.

The third piece is the tools, what the agent can do. Here, it can look up free tables in the booking table, and once the booking is approved, save it there. And the fourth piece is the human checkpoint.

Nothing reaches the guest until someone on your team has looked at it and tapped approve. There are at least three ways to do it. One is to create it manually.

You click on create workflow, and you start adding step by step. So, let's say we start with Telegram message, and here you need to configure the credentials. It's already picked up the credentials of one bot that I created before, but you would need to create also your bot in Telegram and add it here.

Okay, so this step is done. Then we add in the next step, we add in the AI agent, and we also need to configure it. Specifically, we need to pick what model we're using, let's say, Anthropic, Claude Sonnet, to start off, then what memory it needs to use.

Memory, that's where the agents store its notes. That's how it learns, and what it learned, it keeps in this memory. Not to confuse with the booking table that we also will need to add.

So that's simple memory. and the tool and here we need to add this tool storing into the booking table and we'll go with just internal data table good enough to start off you also need to configure the table and many more so that's just how to know how you can update it in future creating it on your own from the scratch like this is too tedious. The second way to do it is to ask n8n to do it, build with AI.

This is the prompt that I created with Claude. So basically, you describe what you want an AI agent to do. Let's see if it will succeed.

Okay, eventually, n8n assistant built the agent and it's different from what we built before. It didn't build workflow. Like here we have a workflow with every step defined and it built a different feature called agent.

I must say it's a much simpler setup but it gives you less control. So importantly, here we can define the rules that the manager must approve and here It just sits in the prompt that asks the manager to approve. That's a promise that the agent does, but it's not guaranteed.

But still, a pretty decent setup. How I set up this workflow is with Claude Code. You don't need to create no Telegram bots, it will do everything for you.

And I'll share my prompt in the comment and my Claude Code setup, and it's really easy. And if you'd like an agent like this working in your business, the link to book an AI audit is in the comments.