AI Transparency

Full transparency — and why this isn't fine print, but the most exciting part of the story.

This portal is built in close collaboration with Artificial Intelligence. I don't hide that — I show it. Here you'll find, out in the open, what comes from me, what was created with AI, with which tools, and who carries the responsibility in the end (spoiler: me). And since this is a learning portal about AI anyway, I'll also tell you how it works — because the way it's made is the best illustration material I have.

What comes from whom

Idea, concept, architecture, design system, decisionsHuman. The foundation — what the portal wants to be, how it looks, what's right and what isn't — comes from me.
ResearchHuman + AI tools like Perplexity. A search engine with a turbo; the selection and judgement stay with me.
TextsMixed. Some I write or dictate myself, some the AI drafts from my substantive brief. Whatever goes live, I have reviewed and approved.
Translations (56 language variants)AI-assisted. 28 languages, each additionally in Plain Language. Translated pages carry a visible notice.
Images (individual ones, mainly in the blog)Partly AI-generated with Google Gemini (internally „Nano Banana“). I label such images.
CodeBuilt for the most part with Claude Code (on a few afternoons, out of sheer curiosity, also with Google's Antigravity). Tests, reviews and architecture decisions are my responsibility.
ResponsibilityEntirely with the site operator. AI here is a tool and a co-worker — not the publisher.

How to spot AI on this portal

  • A translation notice in the footer of every page that doesn't appear in the original language (German) — plus more prominent banners on the home page, interviews and blog.
  • A caption on AI-generated images.
  • This page — linked from the footer and the legal notice.

That's all it takes. Less would be dishonest.

The masterstroke: „This isn't really vibe coding anymore“

I wear the „100% Vibe Coded“ stamp on the home page with a wink — but honestly, it undersells things. What happens here has a more precise name: Spec-Driven Development. Not „I type wishes into a chat and see what comes out“, but: I write the specification — directives, skills, coding and architecture guidelines, concepts for every article and every demo. That's the source of truth. The implementation is handled by a team of AI agents:

  • Dispatchers (virtual project managers) break my intent down into tickets and fill a task queue.
  • Developer, test, architecture, UX and review agents work through the tickets — often in tandem: one builds, one checks. Checks and balances, like in a real team.
  • The whole thing runs in loops: one agent keeps the others moving, monitors progress, starts new work. There's actually a term for it as of 2026 — Loop Engineering: „loops that run while I sleep.“

My role? I hardly write code myself anymore. I set the architecture, design system, concepts and decisions, and I put up the quality gates. The classic junior tasks — typing, translating, testing through, repeating — take care of themselves.

The best part: I don't even have to sit at the computer. Earlier, while I was writing the book, today while gardening or simply watching YouTube — I dictate my requirements. Thanks to speech recognition (Whisper) they land in an inbox, my dispatchers turn them into tickets, and the next morning the features are there.

An example: how one sentence becomes hundreds of tickets

So no false impression arises: I did not write the vast majority of these tickets myself — that's handled by my virtual project managers. An example — I say just one sentence:

„From now on we support Swahili.“

What then unfolds, without me intervening any further:

  1. First the system creates its own working branch and checks it out — a „feature branch“, here feature/add-swahili. Everything that follows happens there in isolation, so the live site stays stable at all times.
  2. A dispatcher reads the request and notes: translation guidelines for Swahili are missing.
  3. It commissions thorough research into idioms, expressions and language peculiarities and gathers it all into a translation guideline.
  4. It checks: are there rules for Plain Language in Swahili? No → so it derives them from German and English.
  5. It sets up review tickets: do structure and quality match the other language guidelines? A test text is translated, a reviewer checks it.
  6. Once the guidelines are solid, the system adds Swahili to the central language list, extends the language switcher in the menu (flag, alphabetical position) and wires up every place in the code that needs to know about the new language — from the SEO locale to the ELIZA chatbot.
  7. Now a script runs that maps out the entire portal — every single page — and creates a translation task into Swahili for each route.
  8. As soon as a worker has translated a page, the system automatically spawns the next ticket: a quality review for exactly that page. automatic
  9. Once all routes are translated and checked, the system builds the language bundles — content, interface texts and the glossary highlighting — so that Swahili actually appears in the browser. automatic
  10. It commits the work and runs a full production build as an integrity check — if that fails, the chain stops and reports back to me.
  11. Once the build is green, the system merges the feature branch back into the main line — and Swahili goes live.

From a single sentence, in one night, hundreds of tickets emerge — commissioned and worked through by a team I never personally brief.

And the system reports back to me

It doesn't just run blindly. When something needs my attention — say a new font that could break the automatic glossary highlighting, or a decision only I can make — it writes me a report in my inbox. If I don't react, it reminds me. I'm not the person who does everything. I'm the person who decides.

Where this leads

I started this portal just as Claude Code had freshly arrived — the first line of this codebase dates to May 24, 2025, two days after the tool's official launch. Since then, 11,837 commits have accumulated, around 30 per day on average, many of them at night while nobody was at the computer. Behind them stand an estimated 10,000-plus completed tickets. One human, one design system, clear rules — and a team of agents that does the rest.

Why I do it this way

Honestly: at this scale, this portal would never have been feasible for me alone. 28 languages, each additionally in Plain Language, built with accessibility in mind, over 100 articles and dozens of interactive demos — that's the work of an entire team. AI is what makes this very inclusion practical: that a single person builds and maintains a free, multilingual educational resource at this depth, designed to be as accessible as possible. Much of it emerges passively, in the evenings, from dictated ideas.

And that is exactly my second goal with this portal: to show how it's done. I'm convinced this way of working holds barely imaginable potential — for teachers and for everyone who wants other people to know more and be able to do more; for inclusion; for access to education at all. If a single person can build this on the side in the evenings — what could schools, associations, dedicated individuals do? That's why this portal exists. Free knowledge, shared.

The tool to build your own

This portal is available as a ready-made starter template, with a design system and tools. With vibecore, an AI agent turns it into your own learning portal, no coding skills needed.

More about vibecore

Where the limits are

AI makes mistakes — confidently and convincingly. That's why no content goes live unchecked: the editorial responsibility lies with me. If you still find an error, let me know (feedback form) — corrections are welcome, not embarrassing. Artificial intelligence here is a tool in human hands, not an autopilot without supervision.

Affiliate link to the publisher

The “Springer Verlag” link on the book page is an affiliate link of the Springer Nature Author Affiliate Program (partner network Awin). If the book is bought through this link in the Springer shop, the author receives a commission from the publisher. Nothing changes for buyers: in Germany the book price is fixed by law and is the same from every source. On click, Awin sets a cookie with a 30-day lifetime in order to attribute the purchase. The links to Amazon and to the local bookshop are not affiliate links.

As of September 18, 2026. This page grows with the portal — and with the EU rules on labelling AI content (AI Act, Art. 50).