Create your first AI agent in an hour: here's how to do it

AI has become daily practice for many communication professionals. Quickly rewriting an email, summarizing a text, or generating a social post — it happens fast and often yields immediate results. This primarily involves working with standalone prompts in tools like Copilot or ChatGPT.

And that is exactly where we are missing opportunities.

Because the real step forward lies in AI agents: digital colleagues that you set up correctly once and that then structurally take work off your hands. During our Lunch & Learn on AI agents, AI specialist Tom Kottink (Loo van Eck) showed how to create your first AI agent in less than an hour — without coding. In this blog, we summarize the key insights for you so that you can get started today.

What is an AI Agent?

Most professionals still use AI like this:

1. ask a question
2. get an answer
3. start over

That works fine for occasional tasks, but quickly becomes inefficient for repetitive work.

An AI agent handles this differently. Think of an AI agent as a digital colleague with a fixed role, working method, and memory. Instead of having to explain what you want over and over again, you define how a task should be performed just once.

Compare it to training a colleague. The first time, you explain in detail what the goal is, what you consider important, and what the end result should look like. After that, you expect them to pick it up independently. That is exactly how an AI agent works.

The major advantage lies in scalability. Once you have created an agent that works well, you can share it with colleagues. This makes your approach not only more efficient for yourself but also usable for your entire organization.

Don't start with AI, but with your work

The biggest mistake many people make is starting with the tool. While the real question is much simpler: which task do you want to delegate?

Think of work you do often and that is fairly predictable. For example, writing social posts, rewriting emails, or summarizing documents. It is precisely these types of tasks that lend themselves well to an agent, because you already know what the end result should look like.

A handy way to clarify this is to pretend you are hiring someone. If a colleague is sitting next to you later taking over this task—what exactly do they need to do? What knowledge do they require, and when will you be satisfied with the result?

Here's how to build an AI agent (in 5 minutes)

Building itself is surprisingly simple. Most AI tools already offer this feature. In Microsoft Copilot, you work with agents, in ChatGPT you build GPTs, and in Gemini they are called Gems. You can also work similarly in Claude via Projects.

The steps are basically the same:

1. You create a new agent (or GPT / Gem).
2. You place your instruction (prompt).
3. You may add examples or documents.
4. You save it and test it.

The challenge, therefore, lies not in building, but in designing. How precise are your instructions? How clear is your goal? And how well have you thought about the way the officer needs to work?

What makes a good AI agent?

The difference between a mediocre and a strong agent almost always lies in the instructions. Many prompts remain too general. For example, it might say: “help with communication.” That sounds logical, but for AI, it is simply too vague. A better prompt is: “write clear and understandable resident letters in B1.”

The more specifically you describe what you expect — think of tone of voice, sentence length, or structure — the better the output will be. Think of it as a briefing to a colleague: make clear what role the agent has, what the goal is, and when the result is good.

It also helps to provide guidance on the working method. Instead of heading straight for a final result, you could, for example, have an agent first ask follow-up questions, then gather context, and only then write. This way, you steer towards quality rather than speed.

Start small and build from there

It might be tempting to build an all-rounder right away. An agent that does everything sounds efficient, but in practice, it rarely works well. When creating an AI agent, it is more effective to start small. Choose one task that you perform often and for which you know exactly what a good result is. It is precisely this clarity that makes it easier to configure your agent properly.

Your first version doesn't have to be perfect. In practice, it is often already largely good (about 80%), and the real gain lies in testing and improving. By working with it, you see where there are still minor issues and can fine-tune it more specifically.

From there, you can expand. Instead of one all-encompassing agent, you build multiple specialist agents, each good at something. Together, they form a digital team that increasingly supports your work.

You remain responsible

Perhaps the most important point to keep in mind: keep thinking for yourself.

AI is designed to be persuasive. That makes it powerful, but also treacherous. The output often sounds good, even if the content is incorrect.

That is why your role remains essential. View the officer as a junior colleague whom you mentor. You provide feedback, expect improvement, and in doing so, build a better result together.

Get started immediately with your first AI agent

Do you want to experience how this works for yourself and start building an AI agent today? In the webinar on which this blog is based, we shared a concrete prompt that allows you to immediately lay a strong foundation.

You don't have to make it complicated. Start with one task, clearly outline your expectations, and build your first version. Test it, refine it, and expand from there.

Download the AI ​​agent prompt and get started right away.