Yandex 360 Explained How to Teach AI Assistants to Work with Internal UI Kits

Yandex 360's frontend team shared how to teach an AI assistant to work with an internal UI kit and design system — and why special repository preparation is necessary without it.

The "wow" effect and its downside

The first experiments with AI assistants usually make an impression: MVP in five minutes, code on the first try, it seems the future is already here. But when it comes to a real corporate project with internal libraries and its own UI kit, the illusion falls apart. The assistant invents components that already exist in the system, ignores established patterns, and violates design system conventions.

Yandex 360 encountered this systematically: 90% of the frontend team uses AI assistants every day. The incompatibility problem with internal infrastructure turned out to be not an exception, but the rule.

What AI-ready repository means

Valery Baranov, Head of Frontend Technology at Yandex 360, explains the root of the problem: public language models were trained on open data. Internal components, corporate libraries, and project conventions didn't make it there. This means the AI needs to be given context explicitly and systematically — not through manual prompts each time.

The team developed several principles:

• Structured documentation of each component accounting for the model's context window
• Hint files (`AGENTS.md`, `.cursorrules`) describing architecture and forbidden patterns
• Explicit examples — both correct and incorrect — for each entity
• UI kit metadata automatically coming into context when requested
• CI/CD description and code conventions in machine-readable format

Design system as machine-readable artifact

A separate task is teaching the assistant to follow design system patterns. Internal components don't make it into model training data, so AI starts hallucinating: inventing non-existent APIs or using outdated patterns.

Yandex 360 added an additional documentation layer. Each component is described not only for developers but also with language models in mind: typical errors, acceptable variants, forbidden combinations. This allows the assistant to generate code that passes design review requirements on the first try.

"We made frontend projects truly AI-ready: taught assistants to understand the structure of our repositories, work with internal libraries, and even follow design system patterns," —

Valery Baranov, Yandex 360.

Infrastructure perspective on the problem

Yandex 360 deals with "common frontend" — unified technical components, shared CI/CD, platforms for distributing components across teams. This makes the task especially critical: if one component is described incorrectly, assistants make mistakes in all projects that use it.

So the team's approach is not point fixes to prompts, but systematic work with documentation and repository structure at the platform level.

What this means

Yandex 360's experience shows: an AI assistant doesn't become an effective tool without investment in context. Teams that make their repositories and design systems machine-readable will get real speed gains — others will continue spending time fixing hallucinations.