# Empowering designers: AI, design systems, and quality | Schema by Figma 2025

Source: https://www.youtube.com/watch?v=jvJ6uQLQ8Fo
Recap page: https://rapidrecap.app/video/jvJ6uQLQ8Fo
Generated: 2025-11-17T14:37:31.634+00:00

---
## Quick Overview

The presentation concludes that empowering designers through AI integration with design systems requires setting and maintaining quality across multiple tools, enabling system exploration and iteration by preventing off-system distractions, and designing to be technology agnostic to ensure system data is queryable regardless of the tool used, which ultimately leads to better collaboration and faster iteration cycles.

**Key Points:**
- The core challenge is setting and maintaining quality across multiple tools (design and engineering) due to the proliferation of LLM tools that developers are familiar with.
- LinkedIn's design system team shifted scope from merely managing libraries to offering a 'Design System as a Service' model, enabling proactive governance across design and code.
- A key learning is the critical need to enable on-system exploration and iteration by preventing off-system distractions, ensuring stakeholders remain focused on system assets.
- The system must be technology agnostic, meaning its data must be queryable regardless of the tool (Figma, Make, etc.) being used, which helps align contributors across disciplines.
- By applying the system assets (like tokens and specs) directly into tools like Figma's 'Make' feature, designers can prototype and iterate faster without copying/pasting external content.
- The team learned that expecting AI/LLM tools to change is necessary; they are now focused on enabling collaboration across multiple disciplines company-wide.

![Screenshot at 00:08: Grant Blakeman introducing the talk 'Empowering designers: AI, design systems, and quality' at the Schema by Figma conference, setting the stage for discussing how LLMs impact design system practices.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-00-08.png)

**Context:** Grant Blakeman, Staff Design Engineer at LinkedIn, presented at Schema by Figma about how the integration of Artificial Intelligence (AI) and Large Language Models (LLMs) is transforming the role of design systems. The talk focuses on the challenges and lessons learned from scaling design quality and consistency while integrating these new AI-powered workflows both inside and outside the company ecosystem.

## Detailed Analysis

Blakeman argues that the rise of LLMs is blurring traditional role boundaries (designer vs. engineer) and creating challenges in maintaining design system quality and consistency across various tools used by different teams. To address this, LinkedIn's Design System team evolved their role to become a 'Design System as a Service,' focusing on proactive governance rather than just library management. A major lesson learned is the necessity of enabling on-system exploration and iteration by strictly preventing off-system distractions, which is critical for maintaining quality when designers are experimenting with AI tools. Furthermore, the system must remain technology agnostic—the system data must be queryable regardless of the tool (like Make or Figma) being used—to ensure consistency across the company. This approach allows designers to easily apply system content (like tokens or specs) to prototypes, leading to faster iteration and better communication between design and engineering partners. The presentation concludes by emphasizing that embracing change and expecting tools to evolve is vital for success in this new environment.

### Introduction

- AI, Design Systems, and Quality: Lingering and blurring of traditional role boundaries
- LLMs changing our roles
- Need to address challenges at a new scale
- Design System as a Solution Service, not just a library of components.

### Design at LLM Scale

- LLMs tend to converge on sameness (like Shadcn/Tailwind/React)
- Need to avoid this visual sameness while retaining brand identity
- System data must be queryable by context (human or LLM)
- System must provide core support for product building.

### Working with LLMs

- Early work used Figma Make to prototype
- Alignment and accuracy became a top request
- System data (specs) must be extractable and readily available for LLM consumption.

### Enabling Collaboration in Make

- Proactive governance requires new approaches to system governance
- LLMs can be guided to produce consistent results
- Goal is to teach tools how to use the system rather than relying solely on designer knowledge.

### What We're Learning

- Set and maintain quality across multiple tools & contexts
- Enable on-system exploration & iteration; preventing off-system distractions is critical
- Enable collaboration across multiple disciplines, company-wide
- Support broader functional backgrounds
- Stay technology agnostic to keep system data queryable by any tool.

![Screenshot at 00:04: Grant Blakeman walking onto the stage to begin the presentation.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-00-04.png)
![Screenshot at 00:16: The speaker introduces himself as Grant Blakeman from the Design Systems team at LinkedIn.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-00-16.png)
![Screenshot at 01:12: Slide highlighting 'Design at LLM scale' and showing UI components fading to gray, illustrating the risk of visual convergence/sameness.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-01-12.png)
![Screenshot at 02:32: Slide detailing the shift in scope: 'Our scope is no longer managing libraries and providing guidance,' implying a service-oriented shift.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-02-32.png)
![Screenshot at 03:26: Mind map illustrating the breadth of the Design System as a Service, connecting core assets \(left\) to LLM-empowered tasks \(right\).](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-03-26.png)
![Screenshot at 06:10: Slide showing code snippets and stating that providing guardrails for LLMs is crucial for working within the system.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-06-10.png)
![Screenshot at 07:41: Slide showing an example template in 'Make' being used to generate three divergent design ideas from one prompt, demonstrating exploration.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-07-41.png)
![Screenshot at 09:59: Slide titled 'Rapid communication with Make,' showing an LLM chat interface responding to a prompt.](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-09-59.png)
![Screenshot at 11:06: Slide summarizing key learnings: 'Set and maintain quality across multiple tools & contexts' and 'Enable collaboration across multiple disciplines, company-wide.'](https://ss.rapidrecap.app/screens/jvJ6uQLQ8Fo/00-11-06.png)