Overview
By Mike Mozg, founder and CEO of Mozgii, Inc. and chair of the board of MozgMind. September 2026, part 1 of the series “How our site began”. Published 11 November 2026. Reading time: 3 minutes.
The short version: an AI writes code and drafts, and a person decides what the site promises to a client. Without this split you cannot assemble a site in a week.
What happened in spring?
The project began on 28 April 2026. In the first four days it got 42 edits: the site frame, the first pages and the build settings. Then work on the site stopped for almost five months.
The frame proved useful. When I came back to the site on 24 September, I did not have to choose technology again. The pages are built in advance on Next.js, the site runs in the Timeweb cloud in Russia, and the texts lie in separate files, not mixed with code. The Yandex Metrica counter works only after the visitor agrees, there is no foreign analytics, and visitors’ data is stored in Russia, as 152-FZ, Russia’s personal data law, requires.
What happened in the week the site came together?
The busiest days were 27 and 28 September, with 110 and 108 edits in a day. The AI wrote the code of pages, placed texts into templates and fixed what the build broke. I set the task, read the result and decided whether to keep it or redo it.
By 28 September the site had a home page, service pages with a fixed price and term in the contract, and the first articles. They covered how much AI adoptionAI implementation. AI implementation is the path from an idea to a working solution: a company chooses a process, connects artificial intelligence to it, trains its people, and measures the result in money. costs and how an agent differs from a chatbot. AI adoption is the path from an idea to a working solution: a company picks a process, connects AI to it, trains people and measures the result in money.
What did the AI do, and what did the person do?
An AI quickly does what can be checked: a page from a sample, a card from a template, a fix from the text of an error. It does slowly and badly what needs a choice: what we sell, to whom and with what guarantee.
So the choice stayed with me. Which pages are needed, which figures we promise, what may be written about clients and what may not. No edit went to the site without my consent, and this rule still holds.
Which rules does vibe coding need?
Vibe codingvibe coding. Vibe coding is a way to build software in which a person describes the task in words, an AI model writes the code, and the person hardly reads the code itself. means writing code with words, not by hand. It works as long as the rules are written in advance. Otherwise each next request to the AI quietly breaks what you agreed yesterday.
Our rules appeared in the first week:
- Visitor counting works only after the visitor agrees.
- Every page has one top-level heading, its own description and a date.
- Texts avoid officialese and stay plain.
The rules are written in the project files, and the AI reads them before every task. Scripts check the build, the metadata and the texts. So a violation is visible at once, not a month later in a search console.
How do you build a website with AI in five steps?
If you want to repeat such a launch, this order worked for us. It does not depend on which AI you use.
- Write what the site sells and promises, with numbers.
- Write the rules that must not be broken, from the law on personal data to how headings are written.
- Pick one sample page and finish it. The AI builds the other pages from it.
- Set tasks one at a time. After each, run the check of the build and the metadata.
- Publish only what you have read yourself.
The AI speeds up every step, but the person sets the order of the steps.
What would I do the same way?
- Start with rules, not with design.
- Set tasks one at a time and check each with a script, not by eye.
- Never give the AI the decisions about what the site promises to a client. The company answers for them, not the model.
Mike Mozg has led software development for 20 years, including for Walmart, Disney, Apple and Microsoft.
Next in the series
Part 2: how an AI writes our articles and why every fact is checked against a source with a date.


