Agentic work · Nijmegen

Agentic work: building with AI agents

I'm Thijs, Head of Product at WeGuide. Almost everything I've built lately, I made together with AI agents. I'm a Cursor Build Ambassador and part of the SpaceSI community. That means I regularly organise workshops and meetups in Nijmegen, like Build with Cursor and Cafe Cursor, to help others work this way.

What is agentic work?

Agentic work means you give an AI agent a task instead of a question. The agent makes a plan, carries it out, tests the result and adjusts. You set the goal, the limits and what counts as good enough.

The difference with a chatbot: an agent works in your own files and tools. It reads your code or documents, runs commands and shows you what it changed. That makes it faster than copying and pasting from a chat window, and it asks for a different way of working.

Agentic coding vs vibe coding: what's the difference?

Vibe coding

You describe what you want, the AI writes code and you check if it works. Fast for a first prototype, hard once it grows.

Agentic coding

You give the agent a plan, rules and tests, and you read every change. A bit more preparation, but it keeps working and stays easy to maintain.

How I work: Cursor, Claude Code and Codex in practice

A steady rhythm helps more than the perfect model. This is how I approach a new feature or prototype:

  1. Context first. An AGENTS.md in the repo that explains how the project works, which rules apply and how to test. Every agent reads it first.
  2. Plan before code. In plan mode the agent writes a plan first. I read it, cut what's too big and only then let it build.
  3. Small steps. One change at a time, with a test or a screenshot as proof that it works.
  4. A second pair of eyes. Another agent or model reviews the work. I decide what goes in.
  5. Work in parallel. Separate tasks go to subagents that work at the same time, while I keep an eye on the product.

These are the tools I use most:

You'll find examples of what I built this way in my portfolio, like a web app that counts your exercises through your camera and a voice agent for GP practices.

Not only for code: harnesses for marketing, sales and design

Many people think agents are only for developers. I use them for almost everything I do. For that I have different harnesses: a fixed setup for each kind of work, with instructions, skills, tools and scripts, so an agent knows how I want things done.

I use those harnesses every day. Code is only part of the work. The same approach works for anything you can describe in steps and check.

Building AI agents with your team: workshops

In a workshop we work on your own work, not on a demo. It suits product teams, designers, researchers and developers who want to find out what AI agents can do for them.

We agree on the length and format together. Email me what you want to achieve.

Meetups and builder community

As a Cursor Build Ambassador I regularly organise meetups in Nijmegen for people who build with AI, like Build with Cursor at StartUp Nijmegen and Cafe Cursor. Anyone can join: build together for an afternoon, show what you're working on and help each other out. New dates are on the Cursor Nijmegen Luma page.

Frequently asked questions

What does agentic mean?

Agentic means an AI takes steps towards a goal on its own: planning, doing something, checking the result and adjusting. You give it a task and limits, the agent carries it out and you review it.

What's the difference between vibe coding and agentic coding?

With vibe coding you let the AI write code and mostly check whether it works. With agentic coding you give the agent a clear plan, rules and tests, and you check every step. The first is fast for a prototype, the second keeps the code easy to maintain in the long run.

How do you learn agentic coding?

By doing it on a real project. Start small, write a plan with the agent first, let it build and test, and read every change. A workshop or meetup helps you spot the pitfalls sooner.

What does an AI agent cost?

To work with agents like Cursor, Claude Code or Codex you usually pay a subscription per person, often around 20 dollars a month for the entry plan. Heavy use costs more. An agent you build yourself mostly costs time and API usage.

Is Claude Code only for programmers?

No. Product people, designers and researchers use it too, for prototypes, data analysis, documents and small tools. You don't need to write code, but you do need to explain what you want and look at the result with a critical eye.

What's the difference between ChatGPT and an AI agent?

ChatGPT answers a question. An AI agent carries out a task: it reads files, runs commands, tests what it makes and keeps going until it's done. You stay in charge and sign off on the result.

Want to join in?

Come to one of the meetups in Nijmegen. Luma shows when the next one is. Would you rather talk about agentic work first? Email me.