The article that filed itself
This article did something unusual the moment it was published: it filed itself.
Nobody copied it into a system. Nobody got a reminder to "update the knowledge base". Seconds after publication it was in the memory of Aidan — our AI guide, whom you can meet in the bottom-right corner of this page — and he could answer questions about it, linking to the source.
That sounds like a small thing. It is the opposite.
The problem: knowledge bases rot
Every company that has tried giving an AI assistant access to "our own knowledge" knows the pattern. You gather the documents, index them, and the assistant gets smart. For two weeks.
Then someone publishes a new page, changes a price, rewrites a description — and the knowledge base doesn't know. The assistant still answers, politely and confidently. Just from the day before yesterday. A knowledge base that is nobody's job to update is outdated from the day it goes live.
The classic fix is a reminder in someone's calendar. It lasts exactly as long as those things last.
The fix: publishing IS the update
We built it as a chain instead, where each link pulls the next:
- Someone publishes a page in the CMS — an article, a case, a product description.
- The CMS calls a webhook — a small message the site itself has registered: "something happened to this page". It's opt-in per site, visible in the admin, and every message is signed so nobody outside can forge one.
- The site's receiver fetches the finished page, strips the menus and chrome, and sends the content to the knowledge base.
- Aidan looks things up in the knowledge base every time someone asks him a question — and cites the page the answer came from, with a link.
No nightly jobs. No calendar. Nobody remembering anything. The day this article went live, it was its own dress rehearsal: publish, webhook, archive — and Aidan could answer.
The boring details that decide whether it works
The hard part of systems like this is never the happy day. It's the edges:
- Drafts never leak. Only published content reaches the knowledge base — a half-finished draft is not knowledge.
- Repeats are politely declined. Publish the same page again without changes, and the knowledge base recognises the content and says no thanks. No duplicates.
- The chain survives a broken link. If the knowledge base is briefly down, Aidan still answers — just without that day's lookup. Visitors notice nothing.
Each of those three properties was tested by trying to break it, not by hoping.
Why this is the point of AI-native
We call our approach AI-native, and this is the shortest demonstration of what the word means: the systems keep each other up to date so people don't have to.
An ordinary website with a chatbot is two systems glued together by a human. An AI-native site is one system, where publishing, archiving and knowledge connect on their own — so the assistant in the corner never speaks from the day before yesterday.
Want to see it work? Ask Aidan about this article. He has already read it.