How Eurecable is made together with an AI
This new application was rewritten from scratch on Rails 8 by a human and Claude working as a pair. No hype here: what exactly the AI does, what stays with people and where its work shows on the site.
The rewrite
The intent came first: every mechanic described in words in one document, the disputed points turned into questions for the owner, the answers into decisions. Only then the code, always the same way: a scenario, a failing test, the minimal implementation, a green run.
Claude writes the code, the tests and the documentation by these rules and under supervision; the old application stays the reference for behaviour, its data is checked against the new one down to zero differences. Decisions about mechanics belong to the owner.
Where the AI shows on the site
- A draft review of an idea, the first response while people have not come yet. It is marked as a draft, the author can hide it, nobody earns energy or iDeal for it.
- Search and similar ideas work by meaning on a local model: texts are not sent anywhere.
- The AI does not grade ideas instead of people, does not grant rewards and makes no decisions about certificates or the network.
Honest about the limits
An AI makes mistakes, as people do; that is why every mechanic has a test and a description, and every review has a living person who may disagree. If something on the site looks wrong, say so: it is the best contribution to the project.