<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Prateek Saini on Medium]]></title>
        <description><![CDATA[Stories by Prateek Saini on Medium]]></description>
        <link>https://medium.com/@prateeksaini_71722?source=rss-461eebedda1d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*FVqsbEv0N5d2ccfFCcG7eg.jpeg</url>
            <title>Stories by Prateek Saini on Medium</title>
            <link>https://medium.com/@prateeksaini_71722?source=rss-461eebedda1d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 04:11:22 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@prateeksaini_71722/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Most Valuable Design Skill in 2026 Is Not What You Think]]></title>
            <link>https://medium.com/@prateeksaini_71722/the-most-valuable-design-skill-in-2026-is-not-what-you-think-e4f0ddce48d5?source=rss-461eebedda1d------2</link>
            <guid isPermaLink="false">https://medium.com/p/e4f0ddce48d5</guid>
            <category><![CDATA[ux-design-skills]]></category>
            <category><![CDATA[future-of-design]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[design-career]]></category>
            <category><![CDATA[ai-in-design]]></category>
            <dc:creator><![CDATA[Prateek Saini]]></dc:creator>
            <pubDate>Sat, 28 Mar 2026 18:40:57 GMT</pubDate>
            <atom:updated>2026-03-28T18:40:57.580Z</atom:updated>
            <content:encoded><![CDATA[<p><em>It’s not Figma. It’s not prototyping. It’s not even AI prompting. It’s something more fundamental.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MyCtK3hFg6DD1EMX59o7KQ.png" /></figure><p>Every few months, a new article circulates about “the skills designers need in the AI era.” They usually list things like: learn to prompt AI tools, understand engineering basics, build your personal brand.</p><p>These are not wrong. But they are missing the thing that is actually separating designers who are thriving from designers who are struggling to stay relevant.</p><p><strong>The Skill Nobody Is Talking About</strong></p><p>Systems thinking applied to AI collaboration.</p><p>Not “using AI tools.” Not “prompting well.” The ability to structure your design work so that AI can amplify it, and to know when structure enables AI output versus when it constrains it.</p><p>Here is the distinction that matters: a designer who uses AI to generate UI ideas is getting a 10 to 20% efficiency boost. A designer who structures their design system so AI can read it and generate production-ready code is delivering a 10x multiplier to their entire engineering team.</p><p>The first designer is using AI as a tool. The second designer is using AI as infrastructure.</p><p><strong>What I’ve Seen This Year</strong></p><p>I have been working with AI-assisted design workflows at Tap Health for over a year. We have connected Claude directly to our Figma design system via MCP. We have watched AI generate React Native components that engineers ship without modification.</p><p>The single biggest predictor of whether AI produces useful output is not the quality of the prompt. It is the quality of the design system the AI is reading.</p><p>A messy Figma file with hardcoded values and generic layer names produces messy, generic code, regardless of how good your prompt is. A clean design system with semantic tokens, documented components, and consistent naming produces code that is architecturally correct, token-compliant, and ready to ship.</p><p>We went from a 27% design system health score to 87% over three months. Our AI-assisted development velocity tripled. Not because we got better at prompting, because we gave AI something coherent to work with.</p><p><strong>The New Design Hierarchy of Needs</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/680/1*A47vLEGvWE175WJ3MJe3mw.png" /></figure><p>I have been thinking about this as a hierarchy. At the base: visual craft, the ability to make things look right. Above that: interaction design, the ability to make things work right. Above that: systems design, the ability to make things scale right.</p><p>In the AI era, there is a new layer at the top: AI-legible design, the ability to structure work so machines can read it, reason about it, and extend it accurately.</p><p>Most designers are excellent at the first two levels. The best designers I know are exceptional at the third. Very few are thinking seriously about the fourth, which is exactly why it is where the leverage is.</p><p><strong>What AI-Legible Design Actually Means</strong></p><p>It means naming layers semantically (what it IS, not what it looks like). It means building token architectures that express intent, not just values. It means documenting components so an AI reading them understands their purpose and constraints. It means thinking about how auto-layout constraints translate to responsive code.</p><p>It means, fundamentally, treating your Figma file as an API that AI will consume.</p><p>This is not a new skill set from scratch. It builds on everything good design systems work has always required. But the incentive structure has completely changed. Design system quality used to improve developer experience gradually and indirectly. Now it directly determines the quality of AI-generated code.</p><p>Clean design system = code that ships. Messy design system = code that creates more work.</p><p><strong>The Career Implication</strong></p><p>If you are a designer trying to figure out where to invest your energy in 2026, here is my honest take:</p><p>Visual craft is necessary but not sufficient. Every designer has it. AI tools are increasingly good at it. It is not your moat.</p><p>Prompting skills are useful but not differentiating. Everyone is learning to prompt. It is becoming table stakes.</p><p><em>Systems thinking that enables AI collaboration</em>. That is the thing that is hard to replicate, hard to automate, and increasingly scarce.</p><p>The designers I most respect right now are not the ones making the most beautiful screens. They are the ones whose design systems are so well-structured that their AI outputs are indistinguishable from human-crafted components.</p><p>Your design system is your multiplier. Invest in it accordingly.</p><p><strong>Where to Start</strong></p><p>If you want to develop this skill, start with an audit. Run an MCP Readiness check on your Figma file. See your score. The number will tell you exactly where to focus.</p><p>Fix naming first. Semantic layer names are the highest-leverage change you can make. They are free, they are fast, and they immediately improve both AI output quality and developer comprehension.</p><p>Build a semantic token layer if you do not have one. Primitive to semantic to component. This architecture makes theming, AI code generation, and long-term scalability all work at the same time.</p><p>Add component descriptions. 2 to 3 sentences per component explaining its purpose, when to use it, and when not to. AI reads these descriptions when generating code. The difference in output quality is dramatic.</p><p>The designers who do this work now will look back in two years and realize they built a career advantage that compounds. The ones who skip it will keep prompting the same messy system and wondering why the AI output still needs so much manual cleanup.</p><blockquote>Your design system is your multiplier. Invest in it accordingly.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e4f0ddce48d5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Design System Is the Prompt]]></title>
            <link>https://medium.com/@prateeksaini_71722/the-design-system-is-the-prompt-79bc3c341457?source=rss-461eebedda1d------2</link>
            <guid isPermaLink="false">https://medium.com/p/79bc3c341457</guid>
            <category><![CDATA[claude-ai]]></category>
            <category><![CDATA[token-design]]></category>
            <category><![CDATA[figma-mcp]]></category>
            <category><![CDATA[design-systems]]></category>
            <dc:creator><![CDATA[Prateek Saini]]></dc:creator>
            <pubDate>Mon, 16 Mar 2026 18:20:57 GMT</pubDate>
            <atom:updated>2026-03-16T18:20:57.158Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How cleaning up our messy Figma file unlocked AI-generated code that actually ships.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LTjIembXQE9Pm2S4OXrF0g.png" /></figure><p>Last month, I ran a design audit on our product’s Figma file.</p><p>The result? 27% design system health score. Inconsistent spacing. Components with hardcoded values everywhere. Tokens that referenced nothing. Layers named “Rectangle 47.”</p><p>I already knew this was a problem. But what I did not realize was how directly it was killing our AI-assisted development workflow.</p><p>Here is what I discovered: the design system is not just a source of truth for designers. When you connect Claude to Figma via MCP (Model Context Protocol), the design system becomes the literal input for code generation. Every component name, every token value, every spacing decision, Claude reads it all before writing a single line.</p><p>Garbage in. Garbage out.</p><p><strong>The Audit That Changed Everything</strong></p><p>I used Wrack’s Figma Auditor plugin - a 6-tab design system quality engine that includes something called an “MCP Readiness” check. It measures how well your Figma file is structured for AI consumption.</p><p>Our score: 27%. The breakdown was brutal but clarifying: 62% of layers had generic names like “Frame 12” or “Group 3.” Token coverage was at 31%, most values were hardcoded. Component descriptions were empty. Auto-layout was used inconsistently. There was no semantic token layer at all.</p><p>To a human developer looking at Figma, this is annoying. To Claude reading the design system via MCP, this is incomprehensible noise.</p><p><strong>The Cleanup: From 27% to 87%</strong></p><p>With the Figma Desktop Bridge plugin running and Claude connected via MCP, I asked it to audit the file and give me a remediation plan. What followed was one of the most productive afternoons I have had in years.</p><p>Claude did not just identify problems - it executed fixes directly inside Figma. It renamed layers using semantic naming conventions. It built a three-tier token architecture: primitive tokens (raw values like #1A73E8) mapped to semantic tokens (color/action/primary) mapped to component tokens (button/background/default). It added descriptions to every component. It standardized auto-layout spacing across the entire file.</p><p>Three hours later, the health score was 87%.</p><p><strong>The Proof: One React Native Screen, First Try</strong></p><p>The real test was the morning after. I opened Claude, connected it to the cleaned-up Figma file, and asked it to generate a complete React Native onboarding screen - using only our design system components and tokens.</p><p>It worked on the first run.</p><p>Not “mostly worked.” Not “worked after some tweaks.” It ran, it matched the design spec, it used the correct token names, it referenced real component variants. The developer on my team looked at the output and said: “Did you write this?”</p><p>No, Claude did. But only because we gave it something coherent to work with.</p><p><strong>What This Means for Design Teams</strong></p><p>We have been thinking about design system quality wrong. For years, the ROI argument for cleaning up your design system was: “developers will be less confused” and “designers will be more consistent.” Both true. Both hard to quantify in a sprint planning meeting.</p><p>Now there is a new argument: a clean design system is the difference between AI-generated code that ships and AI-generated code that wastes everyone’s time.</p><p>When Claude reads your Figma file via MCP, it is not looking at pixels. It is reading a data structure. Component names become class names. Token values become CSS variables. Layer hierarchy becomes component hierarchy. Auto-layout constraints become flexbox rules.</p><p>If that data structure is clean, the output is clean. If it is messy, no amount of prompt engineering will fix it.</p><p><strong>How to Think About Your Design System Now</strong></p><p>Stop thinking of your design system as documentation. Start thinking of it as a structured dataset that AI will consume to generate production code.</p><p>That means: every component needs a name that describes its function, not its appearance. Every token needs to live in a semantic hierarchy. Every variant needs to map to a real state a developer will implement. Every layer needs to tell a story a machine can read.</p><p>The designers who understand this will be the ones who 10x their engineering team’s output. Not by writing code, by giving AI something worth reading.</p><p><strong><em>The design system IS the prompt.</em></strong></p><p>Clean components = clean code.</p><p>Messy layers = messy output.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=79bc3c341457" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Figma + Claude Workflow That Made Me Rethink Everything About Design-to-Code]]></title>
            <link>https://www.designsystemscollective.com/the-figma-claude-workflow-that-made-me-rethink-everything-about-design-to-code-d4c2f3b9e542?source=rss-461eebedda1d------2</link>
            <guid isPermaLink="false">https://medium.com/p/d4c2f3b9e542</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[react]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[figma]]></category>
            <dc:creator><![CDATA[Prateek Saini]]></dc:creator>
            <pubDate>Fri, 13 Mar 2026 19:23:51 GMT</pubDate>
            <atom:updated>2026-03-31T07:37:16.365Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8P8DtaWak_LoDgULFRZVYA.png" /></figure><p>Over this weekend, I tried something unusual. I’ve been experimenting with a Figma + Claude Cowork workflow and I don’t think I can go back to the old way.</p><blockquote>I sat there staring at my screen the first time it worked thinking “wait… that’s it?”</blockquote><p>This isn’t “AI generates a rough prototype you have to rebuild.” This is design-to-production code with real structure. Here’s exactly what’s happening at Tap Health, and how you can replicate it.</p><p>— -</p><h4><strong>The Problem With the Old Way</strong></h4><p>At Tap Health, I design everything in Figma like I normally would: layouts, flows, components, tokens, screens. Nothing changes there.</p><p>But then instead of writing a handoff doc or jumping into VS Code myself, something different happens now.</p><p>Claude connects directly to my Figma file through MCP. It reads my design system. It understands my tokens and styles. It plans the implementation. And it generates actual React components, dropped right into my project folder.</p><p>Components. Pages. Styles. Tokens. Ready to run.</p><p>The designer-developer handoff as we know it? I think we’re watching it evolve in real time.</p><p>— -</p><h3><strong>How It Actually Works: The 3-Stage Pipeline</strong></h3><p>The workflow breaks down into three clean stages.</p><blockquote><strong>Stage 1: Ideation in Figma</strong> (your existing workflow, unchanged)<br>You design layouts and flows, create components, define tokens and styles, and prepare your screens exactly as you always have. The richer your design system, with named tokens and consistent components, the better the output on the other end.</blockquote><blockquote><strong>Stage 2: Claude Cowork</strong><br>Claude connects to your live Figma file via the Figma MCP plugin. It doesn’t just read the visual output; it reads your design system deeply. It understands tokens, component hierarchies, and style relationships. Then it plans the implementation and generates the React components.</blockquote><blockquote><strong>Stage 3: Output: React Files</strong><br>Production-ready components land directly in your project folder. Not a starting point. Not a rough draft. Actual code that uses your real tokens, your real component names, and your real file structure.</blockquote><p>— -</p><h3>Configure It in 6 Steps</h3><p>Here’s exactly how to set this up from scratch.</p><h4><strong>Step 1: Install the Figma Desktop Bridge Plugin</strong></h4><p>Go to Plugins &gt; Development &gt; Figma Desktop Bridge inside Figma. This plugin opens a WebSocket connection that lets Claude read your live Figma file in real time, components, tokens, styles, and all.</p><h4>Step 2: Open Claude Cowork and Connect</h4><p>Launch Claude Cowork on your desktop. In a new session, Claude will automatically detect the active Figma file via MCP. You’ll see it confirm the connection, and from that point it can read your entire design system as you work.</p><h4>Step 3: Point Claude at Your Project Folder</h4><p>In Cowork, select your React project folder as the workspace. Claude reads your existing file structure so it knows where to drop components. No manual file placement.</p><h4>Step 4: Design in Figma as You Normally Would</h4><p>No special naming conventions required to start. Build your layouts, components, tokens, and flows exactly as you always have. The richer your design system, the more accurate the output.</p><h4>Step 5: Prompt Claude with Your Intent</h4><p>Tell Claude what you want built. For example:</p><p>&gt; “Generate the Dashboard screen as a React component using our design tokens. Drop it into src/components/.”</p><p>Claude reads the Figma file, understands the tokens and styles, plans the component structure, writes the code, and drops it into your project folder.</p><h4>Step 6: Review, Iterate, Ship</h4><p>Open the generated files in VS Code. Because Claude used your actual tokens and component names, the code is already aligned to your system. Tweak what you need, then ship.</p><p>— -</p><h3>The Prompts That Work</h3><p>After experimenting with this workflow, here are four prompt patterns that consistently produce clean output:</p><p><strong>For component generation:</strong><br>&gt; “Read my Figma file and generate the [Screen Name] screen as a React component. Use our design tokens for all colors and spacing. Drop it into [src/components/].”</p><p><strong>For full page builds:</strong><br>&gt; “From my Figma file, generate the [Page Name] page with all its sub-components. Map Figma component names to React component names, apply token-based styles, and place files in [src/pages/]. Include TypeScript props where relevant.”</p><p><strong>For token extraction:</strong><br>&gt; “Read the variables and styles from my Figma file and generate a tokens.css file (or theme.ts) with all color, spacing, and typography tokens as CSS custom properties or TypeScript constants.”</p><p><strong>For design system audits:</strong><br>&gt; “Scan my Figma file and compare the design tokens in use against the tokens defined in our tokens.css. List any hardcoded values that should be replaced with tokens.”</p><p>— -</p><h3>The Tools You Need</h3><p>The stack is simpler than you’d expect:</p><p><strong>Figma</strong> for design (your existing workflow, no changes needed).</p><p><strong>Figma Desktop Bridge</strong> is a free plugin in Figma’s plugin ecosystem. It opens the live WebSocket connection that makes everything else possible.</p><p><strong>Claude Cowork</strong> is Anthropic’s desktop agent. This is the piece that connects to Figma via MCP, reads your design system, and generates production code.</p><p><strong>Your React project</strong> is just your existing codebase. Point Claude at any React project folder and it reads the file structure automatically.</p><p>— -</p><h3><strong>What You Actually Get</strong></h3><p>The four outcomes that matter:</p><p><strong>Faster handoff. </strong>No handoff docs, Zeplin exports, or annotation sessions. Claude reads the file directly.</p><p><strong>Consistent UI</strong>. Because Claude reads your actual design tokens, generated components match your design system, not a rough approximation of it.</p><p><strong>Less manual work. </strong>Layouts, spacing, type styles, color, all mapped automatically. You spend time on logic and edge cases, not boilerplate.</p><p><strong>Production-ready code.</strong> Not a rough AI prototype you have to rebuild from scratch. Real structure, real tokens, real component hierarchy.</p><p>— -</p><h3>Still Early. Still Experimenting.</h3><p>I want to be clear: this workflow is still maturing. Not every component comes out perfectly. Complex interactions and animations still need manual attention. And the better your design system architecture (think three-tier tokens: primitive &gt; semantic &gt; component), the better the output quality.</p><p>But the directional shift is real. The gap between “designed” and “built” is closing faster than I expected.</p><p>If you’re not paying attention to this space right now, you’re going to wish you were.</p><p>— -</p><h4>Try It This Weekend</h4><p>Pick one screen from your Figma file. Set up the MCP connection. Ask Claude to generate it.</p><p><em>You’ll stare at your screen thinking the same thing I did.</em></p><p>— -</p><blockquote>I’m a Product Design Lead at <a href="https://tap.health/">Tap Health</a>, building AI-assisted digital health products. I architect design systems, leads cross-functional teams, and pioneered the Claude + Figma MCP workflow now used as the default design process at Tap Health.</blockquote><blockquote>Follow along on <a href="https://www.linkedin.com/in/prateeeksaini/">LinkedIn</a> for more on AI-augmented design workflows.</blockquote><p>— -</p><p>#FigmaMCP #ClaudeCowork #DesignToCode #AIWorkflow #ProductDesign #DesignEngineering #ReactNative #DesignSystems</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d4c2f3b9e542" width="1" height="1" alt=""><hr><p><a href="https://www.designsystemscollective.com/the-figma-claude-workflow-that-made-me-rethink-everything-about-design-to-code-d4c2f3b9e542">The Figma + Claude Workflow That Made Me Rethink Everything About Design-to-Code</a> was originally published in <a href="https://www.designsystemscollective.com/">Design Systems Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>