Coding Agents — A New Way to Work

In a recent workshop, we were introduced to coding agents — specifically Claude Code and Codex. It is a topic I find genuinely fascinating, and I think it represents something groundbreaking in software development. We are talking about a fundamentally different way of working, where writing code yourself is no longer the default.

That shift is already happening, even for us as students. And honestly? It feels strange. There is something a little disorienting about reaching a point where you are not really coding anymore — especially when you chose this field because you wanted to learn how to build things. I am not entirely sure how to feel about it, but I am choosing to embrace it. The idea of facilitating virtual agents that work on your behalf is too interesting to ignore.

For the workshop assignment, we were tasked with “vibecoding” a quiz application based on a meditation knowledge test that our teacher had originally created on paper.

My workflow looked like this:

I started in Claude Cowork, working inside a dedicated project folder I created for the task.
I gave it a description of what I needed and uploaded photos of the paper quiz — questions and marked correct answers included.
I prompted it to extract the content into a structured markdown file with questions and answers.
On top of that, I had it generate a CLAUDE.md file containing a prompt with instructions for the next step.
I then opened Claude Code in the same project folder and simply told it to build the project based on the instructions in CLAUDE.md.

The result was a functional quiz application — built almost entirely through prompting, with me acting more as a director than a developer.

It is a workflow that still feels a little surreal to go through. But there is a certain craft to it — knowing how to describe what you want clearly, structuring the handoff between tools, and guiding the agent toward a good result. Maybe that is the new skill set we are building.

Emil Kriegel