Summarize https://www.atlassian.com/blog/ai-at-work/teaching-ai-to-speak-our-design-language
—-
Here is a summary of the Atlassian article, **"Teaching AI to speak our design language"**:
**The Challenge: Fragmented Context**
Atlassian found that while people were increasingly using AI to design and code, their AI agents were struggling to generate accurate user interfaces. The problem was that Atlassian’s design language lived across fragmented, human-centric sources—like Confluence pages, Figma files, and documentation sites. Because this information wasn't easily readable by machines, AI tools would often hallucinate components, miss accessibility rules, and use outdated patterns.
**The Solution: "Structured Content"**
To give AI the right context, Atlassian restructured their design system into a machine-readable format called **structured content**.
* They broke freeform text down into consistent, lightweight schemas for components, icons, and guidelines.
* These schemas live in TypeScript files right alongside the code.
* This structured data is fed into an MCP (Model Context Protocol) server, giving AI agents immediate, accurate context without forcing them to scan massive codebases or wordy documentation.
**The Results**
By backing their AI agents with structured content via an MCP server, Atlassian saw major improvements in speed, accuracy, and cost-efficiency:
| Metric | Improvement |
|---|---|
| **Speed** | 34% faster task completion on average |
| **Efficiency** | 26% fewer AI tool calls & 16% lower token usage |
| **Accuracy** | Up to 52% more accurate on specific queries |
| **Code Quality** | 11% fewer errors (due to accurate upfront context) |
**Real-World Impact & "Self-Healing" Docs**
This system allows developers using AI tools (like Cursor) to generate complex, error-free UI components using just simple prompts. It also powers a Slack bot that instantly answers team questions about design guidelines.
To prevent the documentation from becoming outdated, Atlassian also built "self-healing" automation. Whenever a team updates a component in the codebase, AI automatically updates the structured content schemas, ensuring both humans and AI are always working from the latest source of truth.