AI-native publishing

Where I started

Earlier areas of ai.watisdis.com were deliberately simple static pages. That made them easy to publish and inspect, but every new page still combined two different problems: what the page should say and how it should look.

Too much freedom in one step

When AI owns both content and presentation, each new page can become another design task. That added freedom makes consistency harder to enforce and couples content revisions to layout and styling decisions.

Narrow the contract

Give AI a narrower publishing contract: produce structured content that passes a schema. Let deterministic code own layout, typography, interaction, accessibility and presentation rules. Then the interface can change without rewriting the content.

The important change is not Astro or the three visual styles. It is moving variability out of AI's responsibility and into explicit, testable system constraints. The pattern applies beyond publishing: constrain the model's output, validate the contract, let deterministic software enforce the rules, and retain human control where the consequences matter.

From page generation to publishing

Before, AI could produce content, HTML and CSS while making layout and design decisions in the same step. In this experiment, AI produces structured content, validation checks the contract, and deterministic presentation code applies the defined rules.

Same information, different interfaces

This page has one English content source and one Finnish content source. Each interface receives the same language-specific ContentModel. Switching the interface changes presentation, not the source content.

Shared contract, two languages, three interfaces

English and Finnish each have one semantic content source governed by the same schema. Editorial, 8-Bit and Sumi consume that shared contract without presentation-specific copies. The same applies to the semantic data behind the diagrams.

A controlled AI publishing workflow

A future experiment could follow a controlled sequence: create structured language content, validate it, build the site, review the result and approve deployment. The agent does not redesign typography, spacing or responsive behavior for each case; those constraints already live in the system.

Current observations

• A narrower AI responsibility can reduce unintended variation in presentation.
• Structured content enables reuse across interfaces.
• Presentation can change without duplicating the underlying content.
• Static hosting does not prevent rich browser-side interaction.
• More automation can coexist with a deliberate human release gate.