Building uniside.org with agents

The challenge

Build the public face of the company: 28 routes, each in English and German, under a strict design system, with hard performance budgets - and a truthfulness contract that outranks everything else, including marketing instinct. The site may not claim a single thing that is not verifiably true on the day it deploys.

The constraint that shaped the build: it had to be produced the way we produce everything, by agents working from a written specification, with a human deciding and reviewing rather than typing.

The approach

The founder's decisions were captured in a master specification - positioning, design tokens, route map, content rules, forbidden vocabulary, budget numbers. Agents executed it in phases: scaffold, design system, content in both languages, routes and SEO plumbing, then a verification fan-out where separate agents reviewed accessibility, SEO, and performance against the spec's gates.

Enforcement is code, not intentions. The repository contains scripts that fail the build if any budget is exceeded, if a route is unreachable in two clicks, if a German page went missing, or if the copy uses forbidden vocabulary - the operating company's legal name, for instance, may appear only in the imprint, the colophon, and the structured data, and the checker knows it. A marketing superlative fails CI the way a type error does.

The German is not a translation pass bolted on at the end: every document exists as an English and a German file, written in the same sitting under the same voice rules, with Swiss orthography enforced by the same checker.

The result

The site you are reading. All routes statically prerendered in both locales; client-side JavaScript held well under the spec's budgets, with a single interactive client island (the Zurich clock in the chrome - the ticker is CSS-only); self-hosted subset fonts; every number on /open carrying its provenance class. The strongest evidence is not this page - it is that every claim on the site can be checked from the outside, which was the point.

This case study is the proof-of-method for the AI agents and autonomous systems services: the same discipline - written spec, bounded autonomy, verification gates, honest reporting - is what we deploy for clients.