Merijn van Wouden, software developer and inventor. Zaandam, NL.
I like problems that ask for a little invention. Interfaces people are glad to open, algorithms that do more with less, research rabbit holes I should probably have climbed out of sooner. The aim is to understand a problem well enough that the solution becomes small.
Most of the work happens in the hours before the first line of code: reading the domain, arguing with the assumptions, hunting for the one sentence that makes half the requirements disappear. What gets built after that should be the easy part, and usually is.
Currently
Legal tech. Lead developer on a platform that walks entrepreneurs through international law. The Hedera network for the records that have to hold, language models for the reading that would otherwise take weeks.
Provenance. Directing a proof-of-originality system on Polygon and Ethereum, so a buyer can check where a product actually came from.
Travel. Lead AI developer on a neural recommender that plans itineraries anywhere in the world.
Live entertainment. Overseeing development at one of Europe's larger booking platforms.
Four domains, one job underneath. Find the shape of the thing, then make it small enough to trust.
Work I take on
Serverless progressive web apps, and the internal tools nobody writes a case study about. Those get opened every morning by the same twelve people, which is exactly why they ought to be a pleasure to use. I am happy on whatever stack a team already runs: JavaScript and TypeScript, Svelte and SvelteKit, React, Node, PHP, Python, SQL and NoSQL, and plain HTML and CSS done properly.
Done properly is not a figure of speech. It means it works with a keyboard, it works on a bad connection, it works for someone using a screen reader, and it does not ship a megabyte of JavaScript to render a paragraph. An accessible page is usually just a correct one.
What I think
The web should be light. Every byte gets downloaded a million times and burned as electricity somewhere out of sight. Performance work is climate work with better tooling, and it is one of the few kinds of engineering where doing less is measurably better.
Accessibility is not a feature request either. It is the difference between building for people and building for the people who happen to resemble you.
On AI I am split, and I would rather say so than perform confidence. It sits inside most of what I build, and it writes alongside me daily; pretending otherwise would be silly. But the fewer lines a program needs, the more elegant and trustworthy it is, and we have just handed everyone a machine that produces lines by the kilometre. I have swum in enough generated code to be uneasy about it. I want to understand every line I ship. That is a slower promise than it used to be, and I intend to keep making it.
I worry about the juniors, too. The puzzles that used to hurt are the ones that built the mind for the job, and now they are solved before anyone gets to struggle with them.
Blockchain gets the same treatment. I take the work where a shared ledger genuinely earns its place, and leave the rest alone.
Background
I studied Artificial Intelligence at the University of Amsterdam (BSc, 2009 to 2014). Machine learning, neural networks, recommender systems, language. Mostly it taught me how to read a paper properly and how to be sceptical of a demo, which has turned out to be the useful half.
Since then AI has moved from subject to instrument: agentic interfaces, recommendation, language processing. Ten years of being unimpressed by demos is a strange qualification to have, and it has never been worth more.
Dutch and English daily, Russian by ear, German and French on paper.
Notes, thoughts as they arrive, newest first.
On first drafts. A first version is a sketch of the problem, not an answer to it. I keep mine long enough to learn what the thing actually is, then throw it away without ceremony.
On leverage. A generated draft is cheap. Understanding is not. So I let the machine do the typing and keep the judgement.
On small things. Most of the difficulty in an interface is deciding what not to build. The rest is arithmetic.
On tools. I have lost entire days to sharpening things that were not blunt. I have also never once regretted a good tool. Knowing which day you are in is the actual skill, and I have not mastered it.
On this page. The text you are reading is a square spiral. It seemed a better use of a homepage than a hero image and three testimonials.
Elsewhere