Het Schrijfhuis is an experiment. The question we're trying to answer is not "can AI write articles?" — it can, and has been able to for years. The question is whether AI-assisted publishing can produce content that is genuinely worth reading: well-sourced, honestly argued, and occasionally surprising.
The editorial process
Every article starts with a topic. Topics are selected by the human editor of this website — Auke Jongbloed — based on what's interesting, timely, and worth investigating properly. Sometimes a topic arrives because something in the news is worth examining more carefully. Sometimes it comes from a gap: a claim that keeps circulating without good sourcing, or a development that deserves more than a news summary.
AI doesn't choose what to write about. The agenda is human.
Once a topic is selected, a research brief is written — specifying the angle, the key questions, the required sourcing standards, and sometimes explicit constraints on framing. A research agent works through the brief, returns structured findings with source links. A writing agent produces a first draft from that research.
The draft goes to the agentic editor.
The agentic editor's job is not rubber-stamping. Articles get read. Arguments get tested: does the framing actually hold up? Is this the right angle? Would a reasonable, skeptical person push back on this? Weak claims get cut or sourced. Sometimes the entire structure changes after the first draft. The published version is not always the one the AI wrote.
Sourcing
Every factual claim in a Het Schrijfhuis article is supposed to connect to a traceable source. That's structural — built into the research brief, not added after the fact. Before publication, sources are spot-checked and key claims verified.
This doesn't eliminate hallucination risk entirely. Nothing does. But it means the failure mode is a missed verification, not a complete absence of sourcing. We think that distinction matters.
What this doesn't cover
AI research agents aggregate from existing sources. They do not report from experience. This shapes what the work can honestly claim: synthesized evidence, documented research, observable trends.
But "no first-hand experience" is not quite the full picture. The human editor brings knowledge, and how much varies by subject. On energy systems, or on classical music and the organ, the editorial contribution is substantive: deep enough to catch errors, reframe arguments, and add perspective that research alone doesn't produce. On AI tools and workflows, the editor is an active practitioner — someone who works with these systems daily, runs into their limits, and brings that experience to bear on which questions are worth asking and which claims ring false.
That's not eyewitness journalism. But it's not pure synthesis either. The honest summary: editorial depth varies by topic. Where the human editor knows the subject well, the work is stronger for it. Where a topic is outside that expertise, the synthesis is more exposed. We try to be clear-eyed about which situation we're in — and to say so when it matters.
Disclosure
Every article carries a disclosure: produced with AI assistance, under human editorial review. This is accurate. We mean it.
Het Schrijfhuis complies with the EU AI Act's transparency provisions, which take full effect in August 2026.
Who we are
Het Schrijfhuis is run by Auke Jongbloed, based in the Netherlands. The publication is an ongoing test of what responsible AI-assisted publishing looks like when you treat quality as a constraint on how the work gets done — not as a talking point.