AI in Design Workflows: What Atlassian, Meta, Tesco, and Faire Are Doing Differently

Large companies aren’t just experimenting with AI in design. They’re building infrastructure to make AI repeatable, governable, and scalable within their creative teams.

What stands out isn’t the use of tools like Cursor, Claude, or ChatGPT — anyone can do that. The differentiator is how these companies structure the context AI receives before executing any task.

The Emerging Pattern: Context Engineering for Design

Analyzing the workflows of Atlassian, Meta, Tesco, and Faire, a pattern repeats: whoever controls the context controls the output quality.

It’s not about which model to use. It’s about which instructions, templates, and constraints to feed that model.

This is context engineering applied to design — and it separates companies that get consistent results from those playing the prompt lottery.

Atlassian: Pre-Coded Templates + Instruction Files

Atlassian’s design team realized early that asking AI to generate prototypes from scratch resulted in wrong components, ignored tokens, and inconsistent code.

Their solution has three layers:

1. Pre-coded templates — Foundational Design System elements (navigation, tokens, icons, buttons) come pre-configured. AI doesn’t modify them; it only works on spaces open to interpretation.

2. Instruction files — Approximately 2,000 lines of custom instructions that guide AI. When it encounters a certain type of element, the file dictates exactly which design system component to use, which variable to apply, which token to reference.

3. Visual calibration — The team feeds their design components to AI and asks what it sees. If it interprets incorrectly, they correct it. This improves screenshot-to-code conversion.

The result: from screenshot to interactive prototype in minutes, with approximately 70% Atlassian Design System compliance on the first pass.

The lesson: AI doesn’t replace the design system — it becomes its consumer. Those with well-documented design systems have a disproportionate advantage in the AI era.

Meta: Adoption Playbooks + Collapsing Roles

At Meta, AI integration in design is happening at every level — but with a clear line separating execution from strategy.

Jhilmil Jain, VP of Monetization Design, describes the current model: AI for generating quick screens and coded components (execution), while user research, product intuition, and strategy remain manual.

Meta is building playbooks — documented sets of instructions and processes for designers to follow when using AI. Early documentation and standardization have been crucial for consistent adoption.

But the most revealing case comes from Zevi Arnovitz, a Product Manager who describes how AI transformed his role. With no technical background, he now:

Designs basic UI
Uses Cursor for vibe coding concepts
Hands off directly to developers
Splits tasks across models (Claude for user stories, Gemini for JSX, another model for code review)

His statement captures the trend: roles are collapsing, and one person needs to do much more.

He’s not eliminating designers — he takes on smaller tasks that previously required coordination between PM, design, and engineering. The cycle dropped from two sprints to two days.

Tesco: Figma Plugins With Real Data

Tesco’s approach is pragmatic: build internal tools that connect design to real data.

Their senior designer developed Figma plugins that connect directly to the website database. When the team needs to populate prototypes, the plugin fetches images, product descriptions, ratings, and prices from the live site — and inserts everything into UI components at once.

The stack: Cursor for vibe coding + Figma MCP server for brand consistency.

Figma’s MCP (Model Context Protocol) server lets AI tools access structured design data directly through the API, instead of interpreting screenshots. This eliminates guesswork and keeps results aligned with the design system.

The lesson: real data in prototypes isn’t a luxury — it’s a requirement for testing design with integrity.

Faire: Internal Chatbot for UX Research

Faire, the platform connecting retailers to wholesale suppliers, faces a research scale challenge: hundreds of thousands of retailers across different regions and use cases.

The solution: an internal chatbot called Fairey that fetches user queries and tickets, allowing designers to ask questions like: “Find support tickets from brands about our Top Shop program in the last six months?”

This works as primary UX research without the cost and logistics of direct interviews.

For synthesizing actual interviews, the team uses ChatGPT with a security layer. Transcripts are inserted with structured prompt templates to extract organized information.

The lesson: AI as an access layer to internal data transforms research speed without sacrificing depth.

What These Companies Have in Common

Four patterns emerge:

1. Structured Instructions > Ad Hoc Prompts

All invest in documenting instructions, templates, and playbooks. Nobody relies on improvised prompts.

2. Design Systems as AI Infrastructure

The design system isn’t just a visual reference — it’s the foundation AI consumes. The more structured it is, the better the output.

3. Clear Separation: Execution vs. Strategy

AI accelerates execution (prototypes, code, data synthesis). Strategy, qualitative research, and product decisions remain human.

4. Real Data From the Start

Prototypes with fictional data mask real problems. Tesco and Faire show that connecting AI to real data from the design phase improves both quality and speed.

Implications for Those Getting Started

If your organization is evaluating how to integrate AI into design workflows, the evidence points to investments in three areas:

Context engineering — Document instructions, constraints, and templates before handing AI tools to the team.

Design systems as a platform — Ensure your design system is machine-readable, not just human-readable. MCP servers and design APIs are the path.

Adoption governance — Playbooks, process documentation, and continuous calibration separate experimentation from production.

AI won’t replace designers. But designers who master context engineering will replace those who don’t.

Based on public workflows from Atlassian, Meta, Tesco, and Faire, with additional research on MCP servers, design systems, and context engineering in 2025-2026.

References

Atlassian Blog: “Turning Handoffs into Handshakes: Integrating Design Systems for AI Prototyping at Scale”
Lenny’s Newsletter/Podcast: “The non-technical PM’s guide to building with Cursor” — Zevi Arnovitz (Meta)
Figma Blog: “Design Systems And AI: Why MCP Servers Are The Unlock”
UX Collective: “How top companies are using AI in their design workflows” — Punit Chawla