Why I stopped writing, and why I started again.

September 24, 2026

Summary

In 2021 I decided I wanted to share what I was learning. That turned into a long detour: a stack of books about writing, a Python package, and more than a hundred hours of HTML, CSS and JavaScript courses. Then ChatGPT arrived and I stopped writing. This post is about why I stopped, and why I am back.

The people I looked up to

Most of what I know, I learned from people who shared their learnings for free. Vincent Warmerdam with Calmcode. Josh Starmer with StatQuest. Jeremy Howard and Rachel Thomas with fast.ai. Grant Sanderson with 3Blue1Brown. Richard McElreath with Statistical Rethinking, both the book and the lectures he puts online.

At some point it felt wrong to keep taking without putting anything back. So I decided to start writing.

The first detour: writing

Once I started paying attention to writing, I got genuinely interested in it as a skill. I read a lot of books about it. One of them, Write Tight by William Brohaugh (2007), made enough of an impression that I turned its advice into a Python package that checks your text for the patterns the book warns about.

It did not make me fast. A single post took me weeks. Write, revise, read it out loud, revise again. At some point I would publish it, not because it was finished but because I had run out of patience. Even then I rarely shared it with a wider audience, because I did not think it was good enough.

The second detour: building the site

I could have used WordPress or some other content management system, but I wanted to build it myself, for two reasons. The first governs most of my choices: I wanted to know how it actually worked. The second was that getting something into production was a big deal for a data scientist back then, and I wanted to get as much experience with it as possible, even outside of work.

So I took two courses by Jonas Schmedtmann, one on HTML and CSS and one on JavaScript, more than a hundred hours combined. Not a framework, not a template, just the raw material. What I liked most is that Jonas thinks about it the way I do. Learn HTML, CSS and JavaScript properly from scratch, and you can use any framework later. You will understand what it does for you, you will appreciate it more, and you always have the lower-level ideas to fall back on.

The first version of the site came out of that: a Flask app with blueprint templates, completely vanilla HTML and CSS, and a little JavaScript for simple dynamic elements. Later I took a course on Astro and Tailwind, which is what this site is built with now. All of it took years.

Then everyone could write

The arrival of ChatGPT paralysed me. Writing was the skill I had worked hardest at. It was the thing I got praised for, and the thing that set me apart from other data scientists. Within a few months it felt like a commodity, something everyone could do. I stopped writing.

Today my job title is Product Owner GenAI. I have a team of my own, and that team uses AI all day, every day. Even so, when I recently pushed them to start writing, I returned most of their drafts with plenty of feedback on how to improve them. I guess in that sense there is still value in what I learned. I do not want to kid myself, though. They got better fast, and I am fairly sure some of them run an “Ernst review” skill by now.

The reason I am writing again is something else. It structures and clears my thinking, and it means I can explain an idea clearly when I am standing in a room with no document in front of me.

Now I sit in a park and talk

Five years ago, publishing a single post meant dozens of hours of courses first, and then weeks of revising. I am dictating this post from a bench in the park, on my phone, with my earbuds in, talking into Wispr Flow while my thoughts run wherever they want. Claude takes what I say, structures it, and turns it into a post.

That is the short version, though. It starts with a vague idea that I talk through, and Claude gives it a first structure. Then I leave it. While I am walking, exercising or just thinking, more comes to me: a detail I should add, a point that deserves more weight. I talk that through as well, and the post takes another step. It is far from a straight line, and the first draft did not sound like me either. I sent it back with plenty of feedback, much like the drafts my team sends me.

The inspiration still has to come from me. What has changed is that I can do something with it the moment it arrives, because I only have to say it out loud. It is still hours of work, but a fraction of the time a post used to take.

Was it a waste of time?

If a tool can now do much of what took me years to learn, it is fair to ask what those years were worth. To answer that honestly, I have to look at my work as a data scientist, not just at the writing. What set me apart there was that I liked hard problems. I would happily spend days digging into an ambiguous bug that nobody else wanted to touch, and I learned an enormous amount doing it. Colleagues valued that, and so did I.

However, the customer never saw any of it. From where they stood, the same thing took longer and cost more people. The part I was proudest of was, to them, overhead.

That is the uncomfortable part of what is happening now. A lot of that work is disappearing, and with it goes the time it used to consume, which leaves room for what actually mattered to the customer all along. I get the feeling that the skills I spent years on are losing value quickly, which is difficult to accept, but I have found my peace with it.

I have less time than I have ever had, and I am still getting more done than when I had weeks to spend on a single post. The ideas were never the bottleneck. Everything between having an idea and sharing it was, and that is exactly the part that has changed.

No sponsored or affiliate links. If something here is new to you, the link is an easy way in.