A human is
always responsible.
We build with AI. We do not publish what it produces without reading it. Every change to our code, every page of documentation, and every blog passes a person before it goes out. The human stays responsible. The AI stays a tool.
The rule
IBM wrote this on a training slide in 1979. It has not aged a day.
“A computer can never be held accountable, therefore a computer must never make a management decision.”
We are a human-in-the-loop organisation. That is one rule, and it applies everywhere.
Nothing we produce is published on the say-so of a model. Code, documentation, blogs, research notes: a person reads the output, judges it, and decides. If nobody checked it, it does not ship.
Accountability is the whole reason. A model cannot answer for a claim. It cannot be sued, fired, or asked what it was thinking. The person who published it can, and is. When you read something on this site, someone at Conduction stands behind it.
Where we use it
Four places. Each one produces a draft, and a person takes it from there.
We write code with AI agents.
Our pipeline is called Hydra. An agent writes a change, a second agent reviews it, and a third checks it for security problems. Every change then runs through the same quality gates as any other pull request. A developer reads the result and merges it, or sends it back.
Read the Hydra series in the academy →We draft documentation with skills.
Two skills do most of the work. The journeydoc skill turns a real click-through of an app into a tutorial, with screenshots captured from the running product. The tutorial skill drafts the surrounding explanation. Both produce a draft. A person checks it against the product before it ships.
Browse the academy →We use AI to find and read sources.
AI is good at gathering material and summarising it. We use it to find standards, tenders, prior art, and documentation we would otherwise miss. We check the sources it returns. A summary is a starting point, not a citation.
We use AI to check our writing.
Spelling, grammar, and clarity. Most of our team writes in Dutch and publishes in English as well. AI helps us catch the mistakes that come with that, and tightens sentences that run too long.
Blogs and articles
We are developers. Geeks and nerds, at core. We write code, not blogs.
Good writing means putting yourself in the reader's head. Being thoughtful. Empathic. Building a model of another person's mind so you can predict what they will feel about a sentence before they read it.
Written down like that, it is also roughly the job description of a sociopath. We are fine with this.
The honest problem is narrower. We have thoughts worth sharing and we are not natural writers. AI is how the thought gets out of the head and onto the page in a shape someone else enjoys reading. For a person with something to say and no gift for saying it, that is not a convenience. It is an irreplaceable tool.
The value is not volume. Anyone can generate an endless amount of content, and most of it is slop. The value is the opposite of volume: one specific idea, from one specific person, made readable.
So the answer to the question behind this section, did a person write this or did a machine, is that a person wrote it.
The idea comes from a human. So does the subject, the angle, and the scope we set for the piece. The point being made is a human point. The opinion in it is a human opinion. If an article argues something, someone here believes that argument and put their name on it.
The AI does three jobs, and they are all support jobs.
The AI is a research tool and a writing tool. It is not the author.
The editing is human. A person decides what goes in, what comes out, and what the piece finally says. That is the job we do not hand over.
Where this sits in the AI Act
The EU AI Act's transparency rules apply from 2 August 2026. People keep asking which banner we owe them, so here is the plain answer.
Article 50(4) covers AI-generated or AI-manipulated text published to inform the public on matters of public interest. Two things put us outside it, and either one is enough on its own.
We do not publish on matters of public interest. We write about our own software, how we build it, and how to run it. That is a company writing about its own products, not journalism or public information.
And the paragraph exempts content that has been through human review or editorial control, where a person or organisation holds editorial responsibility. Final editing here stays in human hands, and Conduction B.V. holds that responsibility.
So Article 50 does not apply, and we are not obliged to put a banner on anything.
We wrote this page anyway. Not to satisfy a rule, but because you should be able to see where the machine sits in our process and where the human hand is.
Article 50(1) covers AI systems that interact with people directly: those systems have to make clear that you are talking to a machine. This website has no chatbot and no AI assistant. Nothing on it talks back. Where our products use AI, that is documented with the product rather than here.
Article 4 is the one that does apply to us. Since 2 February 2025 it asks anyone deploying AI to make sure the people using it on their behalf actually understand it. Not a certificate, a working understanding: what the tool is good at, where it fails, and when to distrust it.
We treat that as a hiring and training matter rather than a paperwork one. Everyone who ships work here has used these tools long enough to know their failure modes, and knowing that a model will produce a confident, fluent, wrong answer is exactly why a person reads everything before it goes out. We also write our own training material in public: the Hydra series in the academy is how we build with AI, published so anyone can check our reasoning or copy it.
AI at Conduction.
Is anything on this site written entirely by AI?
No. Everything published here has been read and approved by a person. Some pages started as an AI draft, and the academy tutorials in particular are drafted with skills that capture the product as it actually runs. A person then checks that draft against the product and edits it. Nothing goes out unread.
Who is responsible if something is wrong?
We are. A model cannot hold responsibility for a claim, so we do not let it. The person who approved the text or merged the code is accountable for it, and Conduction is accountable as a company. If you find an error, write to [email protected] and we will correct it.
Do AI agents merge their own code?
No. Agents open pull requests and review each other, which catches a lot. The merge is a human decision, and the change runs the same quality gates as any other pull request: linting, static analysis, tests, and a security review.
Do you use our data to train models?
No. We do not train models on customer data. What we process, and why, is set out in our privacy policy.
Does the EU AI Act require you to label these pages?
No. Article 50 does not apply to us, for two separate reasons, either of which is enough. We do not publish to inform the public on matters of public interest: we write about our own software. And final editing stays in human hands, which the article exempts explicitly, with Conduction B.V. holding editorial responsibility. So no banner is owed. We publish this page regardless, because you should be able to see where AI sits in our process and where the human hand is.
Which parts of a blog are the author's own?
The idea, the subject, the angle, the argument, and the conclusion. Those are human every time. AI contributes source material, a spelling check, and suggestions on phrasing. The author decides what to keep.