Frozen archetypes: a whole page group in seconds, with no code generated

Building a group of pages — a news feed, a blog, a documentation section — is mechanical but large work. Until now an AI coding agent wrote every file by hand, which is slow, spends tokens, and drifts from the standard whenever the model improvises. The Agentic Engineering Infrastructure replaces that with a frozen archetype: a ready-made "project in a box" that an agent thaws into your site with a few parameters.

What a frozen archetype is

Think of it as a cookie cutter for a page group. The frozen archetype is an inert tree of template files with no fixed records — the specifics (which group, which languages, the label, how many examples) are parameters, not content. You say "I want a news section"; the agent presses the cutter with your parameters and out comes the working pages.

It lives in a closed store on your own server, so nothing leaves your machine and every deployment already has it.

How thawing works — file copy, not code generation

Thawing is deliberately dumb, and that is the point. The agent copies the template files into your project and fills in the blanks: it installs the shared content engine only if your site does not already have it, creates the section and a couple of example articles, and translates the section label ("News", "Новости", and so on) into each of your languages. The example articles inherit the ideal page structure — search-engine metadata, breadcrumbs, a table of contents, an FAQ — but use placeholder text and a placeholder image you replace later.

• No model writes code — it copies files and substitutes tokens.
• Any AI model (the built-in brain or any coding agent) produces an identical result.
• The pages are static and work with JavaScript switched off.

When it fits, and when the agent refuses honestly

One archetype does not serve every kind of site. Today there is one — a content collection — and it covers news, blogs and documentation. Ask for one of those and the agent thaws it. Ask for something it cannot serve well — a shop with a cart and checkout, a course with graded tests — and the agent tells you so honestly and offers to build a new archetype for it, instead of forcing a bad fit or inventing fragile code. This same idea — translating multilingual content the scalable way — is described in our note on multilingual content architecture.

Why this matters for an agentic engineering workspace

The whole promise of agentic engineering is that the boring, repeatable parts of building software become a single instruction. Frozen archetypes turn "add a section to my site" from a careful coding task into a request anyone can make — and the result is the same, every time, on your own server.

I want building a website to feel like talking, not like assembling furniture. A frozen archetype is one more place where you say what you want and it simply appears — correct, yours, and on your own machine.