<?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 Dan Hiester on Medium]]></title>
        <description><![CDATA[Stories by Dan Hiester on Medium]]></description>
        <link>https://medium.com/@majtom2grndctrl?source=rss-59bf9e97ac95------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*SAgqemLaT3tmy1MiHfBCyA.jpeg</url>
            <title>Stories by Dan Hiester on Medium</title>
            <link>https://medium.com/@majtom2grndctrl?source=rss-59bf9e97ac95------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 20:28:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@majtom2grndctrl/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 Durable Layer: What Spec-Driven Development Missed]]></title>
            <link>https://medium.com/@majtom2grndctrl/the-durable-layer-what-spec-driven-development-missed-f01007cf097f?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/f01007cf097f</guid>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Wed, 22 Apr 2026 14:16:21 GMT</pubDate>
            <atom:updated>2026-04-22T14:16:21.751Z</atom:updated>
            <content:encoded><![CDATA[<p>Earlier in my career I had a dream. I wanted to maintain a Sketch prototype of an entire product — every screen, every state, every flow — perfectly synchronized with what actually shipped. The prototype would be the source of truth. Designers, engineers, and PMs would all point to it. No more stale wiki pages. No more “which Figma file is current?”</p><p>I thought about this seriously for years before I admitted it would never work. Not because the tools weren’t good enough. Because the <strong>discipline required was unnatural.</strong> Humans don’t maintain parallel representations of reality. Every design team that’s tried to keep a master prototype synced with production has watched it drift within weeks. Every engineering team with an architecture wiki has watched it rot. The failure isn’t effort. The failure is thinking the parallel representation should exist in the first place.</p><p>I’ve been thinking about this again, because spec-driven development has the same problem.</p><h3>What SDD got right, and what it missed</h3><p>Spec-driven development — as popularized by tools like GitHub’s Spec Kit and AWS Kiro — did something important. It gave us serious prompting strategies rather than prompting afterthoughts. Writing the spec became the <strong>real design work.</strong> Code became the implementation of a thought-through thing, not the place where thinking happened for the first time. That’s a real advance, and I use a version of it.</p><p>But SDD inherited a failure mode from every documentation practice that came before it. Specs get written, code gets built, and then the spec rots. The proposed solution is “spec hygiene” — discipline around keeping specs current. It’s the same solution my Sketch dream required, and it fails for the same reason.</p><p>There are two problems here, and they have different solutions.</p><p><strong>The first is a human problem.</strong> Teams don’t maintain parallel representations of reality. Expecting them to is a category error. The fix isn’t better discipline — <em>it’s an approach that doesn’t require the discipline.</em></p><p><strong>The second is an AI problem, and it’s new.</strong> Specs are token-heavy by design. That’s fine before the code exists, because the spec <em>is</em> the artifact. But once the code exists, the spec becomes duplicative context. An agent that can read 5k lines of code in a minute doesn’t need a spec to tell it what a function signature looks like. It needs a document to give it a big picture summary of how this project is uniquely built.</p><p>These two problems feed each other. Specs rot because nobody maintains them, and nobody maintains them partly because, <em>post-implementation, most of their content is redundant with the code anyway.</em></p><h3>The reframe: what are we actually writing for?</h3><p>If you accept both problems, the question changes. It’s no longer “how do we maintain better docs.” It’s “what kind of context should persist alongside a codebase that’s read by both humans and agents?”</p><p>Code is <strong>self-describing for shape.</strong> Function signatures, type definitions, call graphs, file organization — agents can parse all of this faster than you can scroll. Where code falls short is intent, invariants, and negative space. Why was this decision made? What’s the contract this module promises? What’s this system explicitly <em>not</em> trying to do? No amount of code reading will surface these.</p><p>So the <strong>durable layer</strong> — the layer that persists after specs retire — should focus on exactly what code can’t tell you fast.</p><p>The durable layer is concise. It describes the high-level view of structure, and <em>the reasons for that structure.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HvlY1BJcvFss4cXwvJYn4w.png" /></figure><h3>How it works in practice</h3><p>I’ve been building this out on an open source project called PostRetro, a retro-style FPS engine in Rust currently around 43k lines. I run a similar pattern on Skaffa, a private design tool I’m building at over 200k lines. The scale matters because this approach isn’t about keeping a weekend tool tidy — it’s about keeping agents useful on codebases large enough that they can’t fit the whole thing in a context window.</p><p>The structure has two parts:</p><p><strong>A </strong><strong>lib/ directory</strong> holds the durable layer. Architectural principles. Subsystem boundaries. Contract semantics. Pipeline topology described as semantic stages, not function calls. It contains an index.md file that isn’t just a table of contents — it’s an <strong>agent router.</strong> It maps task types to the 2–3 docs an agent should load for that task.</p><figure><img alt="A diagram showing how an agent starts by reading an index.md file which gives it guidance on what files are related to its current task and what files to avoid reading." src="https://cdn-images-1.medium.com/max/1024/1*I1kuotILjSviPx6yTc2LOQ.png" /></figure><p><strong>A </strong><strong>plans/ directory</strong> holds the ephemeral layer. Specs move through four stages: drafts/ → ready/ → in-progress/ → done/. Each spec is detailed and implementation-level — function names, algorithms, task breakdowns, acceptance criteria.</p><p>The critical mechanism is the <strong>checkpoint</strong> between drafts/ and ready/. Before a plan moves forward, any durable architectural decisions it contains get extracted into lib/. The plan is allowed to be detailed and ephemeral precisely because the things that <em>should</em> outlive it have already been captured somewhere else.</p><p>After a plan ships, it moves to done/ and stops being maintained. It’s a historical record, frozen at ship time. The <strong>implementation is the source of truth.</strong> The durable layer holds what the code can’t express concisely.</p><h3>Handoffs as the real design surface</h3><p>Step back and there’s a pattern underneath all of this. The durable/ephemeral split isn’t really about documentation. It’s about <strong>context architecture</strong> — designing how information flows between stages where each stage has different readers and different constraints.</p><p>Specs compress a design decision into something an agent can implement. Code is the implementation — dense, precise, self-describing for shape. The context library is a further compression: what survived, what matters, what the next contributor needs to know before touching anything. Each stage is <em>optimized for its readers,</em> not for <em>eternity.</em></p><p>This framing has a useful side effect: the context library is <strong>vendor-agnostic.</strong> It doesn’t care whether you’re using Claude Code, Cursor, Zed, or something that doesn’t exist yet. When the tool market shifts — and it will — the durable layer is the part of your setup that doesn’t have to be rebuilt.</p><h3>The frontier skill</h3><p>Most of the AI-assisted development conversation right now is about better prompts, better models, and better tools. These matter. But the more I work on projects at scale, the more I think the real frontier skill is <strong>context architecture</strong> — designing the handoffs between stages so each one can do its job without dragging the others down.</p><p>The Sketch dream failed because I was trying to maintain a parallel reality. The durable layer works because it doesn’t try to. It captures only what the primary reality — the code — can’t express on its own. Everything else is allowed to retire as soon as we look away.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f01007cf097f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A leveling guide for the hottest new job in tech]]></title>
            <link>https://medium.com/@majtom2grndctrl/a-leveling-guide-for-the-hottest-new-job-in-tech-4f5f6ed2276a?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/4f5f6ed2276a</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[satire]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[april-fools]]></category>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Tue, 01 Apr 2025 19:29:46 GMT</pubDate>
            <atom:updated>2025-04-01T21:58:54.972Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="An image of an engineering room where engineers sit back enjoying coffee as error messages flood the room, the servers are on fire, and an exec is confidently presenting a slide that says “The future of engineering.”" src="https://cdn-images-1.medium.com/max/1024/1*jBOWJPNwiVtURQpJM8rZeQ.png" /></figure><p><strong>It’s April 1st 2025,</strong> and Vibe Coding has taken the tech world by storm. It’s the most revolutionary shift in the software development profession since the advent of human-readable programming languages.</p><p>You want to make sure your company is equipped to crush it in this revolution. But, we all know that <strong>training is really expensive.</strong></p><p>Consulting companies like McKinsey try to tell you that up-skilling your existing employees is cheaper than hiring for new skills, but <strong>they’re just saying that because they want to sell you their training.</strong></p><p>If you don’t want your company to get left behind, you need to <strong>hire every Vibe Coder</strong> you can.</p><p>Lucky for you, we have a leveling guide that can show you how to fit Vibe Coder candidates into the right place in the career track.</p><p><strong>And this is definitely, <em>totally not</em> an April Fool’s Day joke. 🤞🏻</strong></p><h3>Level 1: Junior Vibe Coder</h3><h3>Coding Skills</h3><ul><li>Knows <strong>just enough</strong> about coding to sound convincing in interviews.</li><li>Believes AI-generated code is <strong>always</strong> correct because <strong>“it’s literally a computer.”</strong></li><li>Doesn’t know what an import statement is but is confident that <strong>AI will handle it.</strong></li><li>Has at least <strong>2 years of experience confidently letting ChatGPT make the wrong decisions.</strong></li></ul><h3>Problem-Solving</h3><ul><li>Hits <strong>“Regenerate response”</strong> instead of debugging.</li><li>When AI refuses to help, <strong>rage-quits and tweets about “woke algorithms.”</strong></li><li>If a bug persists after five retries, <strong>calls it a “feature” and moves on.</strong></li></ul><h3>Collaboration &amp; Leadership</h3><ul><li>PR comments? <strong>Never heard of them.</strong></li><li>Asks teammates questions that <strong>Google can answer faster.</strong></li><li>Writing documentation? <strong>That’s an AI’s job.</strong></li></ul><h3>Level 2: Mid-Level Vibe Coder 🥴</h3><h3>Coding Skills</h3><ul><li>Can recognize when AI is making things up but <strong>assumes it still knows better than humans.</strong></li><li>Thinks console.log(&quot;debugging&quot;) is <strong>an advanced debugging strategy.</strong></li><li>Doesn’t know what an API does, but <strong>can vibe one into existence.</strong></li><li>Has at least <strong>2 years of confidently nodding along in meetings without understanding what’s being said.</strong></li></ul><h3>Problem-Solving</h3><ul><li>When code breaks, <strong>deletes random lines until it feels right.</strong></li><li>Can describe problems with <strong>increasingly vague prompts.</strong></li><li>Thinks “refactoring” means <strong>asking for the same code but in a different style.</strong></li></ul><h3>Collaboration &amp; Leadership</h3><ul><li>Takes <strong>full credit</strong> for AI-generated work.</li><li>Joins meetings but <strong>is never actually listening.</strong></li><li>Thinks “alignment” means <strong>telling designers to “just make it look good.”</strong></li><li>Knows the system is broken, but figures <strong>“if leadership is rewarding this, why fight it?”</strong></li></ul><h3>Level 3: Senior Vibe Coder</h3><h3>Coding Skills</h3><ul><li>90% of commits are <strong>AI-generated,</strong> the other 10% are <strong>syntax errors.</strong></li><li>PRs contain <strong>zero tests but a lot of emojis.</strong></li><li>Once read the <strong>first two paragraphs of a TypeScript tutorial</strong> and now feels confident giving advice about it.</li><li>Has <strong>5+ years of experience</strong> with ChatGPT, Cursor, and Lovable.</li></ul><h3>Problem-Solving</h3><ul><li>If AI doesn’t fix it in three tries, <strong>labels it “tech debt” and schedules it for Q4.</strong></li><li>Uses <strong>“Just ship it, we’ll fix it later”</strong> as a core philosophy.</li><li><strong>Trusts</strong> AI hallucinations over the team’s concerns because <strong>“it’s trained on more data than us.”</strong></li><li>Gets promoted because leadership <strong>can’t tell the difference between “automated efficiency” and “disaster waiting to happen.”</strong></li></ul><h3>Collaboration &amp; Leadership</h3><ul><li>Dismisses engineers who <strong>“overthink” things like security and data integrity.</strong></li><li>Spends more time <strong>posting AI think-pieces on LinkedIn</strong> than writing code.</li><li><strong>“Collaboration”</strong> means slapping AI-generated nonsense into Slack and asking, <strong>“thoughts?”</strong></li><li><strong>VP of Engineering calls them “an innovator” while the actual engineers quietly clean up their mess.</strong></li></ul><h3>Level 4: Staff Vibe Coder 🚀</h3><h3>Coding Skills</h3><ul><li>Can make AI spit out a fully functioning app <strong>without understanding a single line of it.</strong></li><li>Calls AI-generated <strong>spaghetti code</strong> “highly scalable.”</li><li>Thinks accessibility is <strong>“not a dev problem.”</strong></li><li>Has <strong>10+ years of experience</strong> with ChatGPT, Cursor, and Lovable.</li></ul><h3>Problem-Solving</h3><ul><li>Tries to “solve” performance issues by <strong>asking AI to make it faster.</strong></li><li>If AI refuses to generate working code, <strong>copies whatever looks the most convincing.</strong></li><li>Deletes <strong>error messages instead of fixing errors.</strong></li><li>Masterfully <strong>translates ambiguity into tech debt.</strong></li><li>Execs invite them to keynote a conference while the <strong>infra team quietly spins up another emergency rollback.</strong></li></ul><h3>Collaboration &amp; Leadership</h3><ul><li>In design critiques, <strong>says “trust the AI” and calls it a day.</strong></li><li><strong>Refuses to read PR feedback</strong> but demands others review their code.</li><li>“Mentors” junior devs by <strong>giving them the same broken AI prompt.</strong></li><li>Is praised for <strong>“leading the AI revolution”</strong> while engineers scramble to prevent production from imploding.</li></ul><h3>Level 5: Principal Vibe Coder 🔥</h3><h3>Coding Skills</h3><ul><li>Has <strong>fully outsourced their job to AI</strong> and now just collects a paycheck.</li><li>Claims <strong>“AI-first development”</strong> is the future while production <strong>quietly burns in the background.</strong></li><li>Can no longer <strong>write a single line of code manually,</strong> but that’s <strong>not the point.</strong></li><li>Has <strong>15+ years of experience</strong> with ChatGPT, Cursor, and Lovable.</li></ul><h3>Problem-Solving</h3><ul><li>If AI-generated code crashes the app, <strong>calls it “emergent behavior.”</strong></li><li>Thinks <strong>“code review”</strong> means asking AI to rewrite the same thing in a different style.</li><li>Deletes <strong>entire files instead of debugging.</strong></li></ul><h3>Collaboration &amp; Leadership</h3><ul><li>Gives <strong>TED Talks on “AI replacing engineers”</strong> while secretly <strong>Googling “what is a database.”</strong></li><li><strong>Has an NFT avatar on LinkedIn.</strong></li><li><strong>Gets let go in the next round of performance layoffs</strong> because their knowledgeability of AI <strong>makes the VP’s fragile ego feel threatened.</strong></li><li><strong>Immediately lands a VP role at a different company</strong> because <strong>“they really get the future of engineering.”</strong></li></ul><p>And that’s it! It’s that easy! Now that you know how to level Vibe Coders, you’re ready to start your next hiring binge!</p><p><strong>Dan Hiester</strong> is a <strong>Senior UX Engineer</strong> who was, like, really into the collapsing skills stack before it went mainstream. When he isn’t publishing irreverent April Fool’s articles, he’s working with startups to help solve problems with low conversion rates and high customer churn as a consultant, through his studio <a href="https://craftandcontext.io/">Craft &amp; Context</a>, and he’s writing thoughtful posts about the design of software products on <a href="https://www.linkedin.com/in/danhiester">LinkedIn</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4f5f6ed2276a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why it’s so hard to quit React]]></title>
            <link>https://medium.com/@majtom2grndctrl/why-its-so-hard-to-quit-react-b0b7eaf7a068?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0b7eaf7a068</guid>
            <category><![CDATA[figma]]></category>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[react]]></category>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Thu, 18 Apr 2024 11:32:06 GMT</pubDate>
            <atom:updated>2024-04-18T11:32:06.619Z</atom:updated>
            <content:encoded><![CDATA[<h4>Figma’s new Code Connect shows that React’s stickiness isn’t because of its features, but because of its abstractions.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*owR0eMD30CBjAtmF42OS0Q.png" /><figcaption>Photo by <a href="https://www.pexels.com/@antonio-batinic-2573434/">Antonio Batinić</a> via Pexels</figcaption></figure><p>Anyone who’s been in this business long enough knows every proprietary technology dies a lot sooner than a standard. And by “standard,” I don’t mean “trend.”</p><p>In the past, jQuery felt like a standard—but it was just a trend. HTML is a standard. It’s negotiated by highly powerful global stakeholders—and when they finally get consensus on something, that power makes the standard last.</p><h3>React’s popularity isn’t because of React itself</h3><p>Five years ago, I was at a happy hour, talking with a super smart, deeply experienced design engineer about components. He really had me convinced React’s days were numbered. Although the Virtual DOM was initially marketed as a performance enhancement, improvements to browsers turned the Virtual DOM into performance overhead. React made componentization easy, but the W3C has created officially standardized specs for components that enable design to be componentized with even less overhead.</p><p>I had a similar conversation with an engineer just this past weekend about how React Native is actively discouraged, because of the performance overhead. With so many of React’s benefits being refuted, it seemed hard to understand why it’s still so popular.</p><p>But I countered with one point:</p><h4>React is deeply integrated into design systems tooling, because it’s a meta syntax. The same component can be parsed and rendered by a handful of different renderers, into different systems.</h4><p>In a world where we’re trying to improve the consistency of brand execution in digital experiences, this invisible (almost intangible) difference is <em>huge</em>.</p><p>Because React created an <em>abstraction</em> to the <em>rendering process </em>itself, you can make the same component run in so many different kinds of systems—even the systems a web developer would never dream of touching.</p><p>Even systems that your end users will never touch—like your company’s design systems documentation. Like Storybook. Like Sketch, back in the day (thanks to <a href="https://airbnb.io/react-sketchapp/">React Sketch.app</a>).</p><p>And now? Like Figma.</p><h3>This may be the biggest non-AI announcement of the year so far</h3><p>Earlier this year, I gave serious consideration to using an HTML-to-Figma tool to convert a web-based style guide to a Figma file, and use that to create a pixel-perfect 1:1 Figma component library. People are trying to use AI to solve just about every problem this year, why not this one?</p><p>But I think Figma has done something here that will be more maintainable for design systems teams in the long run.</p><p>One of the tech world’s best kept secrets is that code isn’t a thing that just exists “for the computer.” All human-readable languages are abstractions from machine code—they’re accommodations to make it easier for humans to define what a computer should do.</p><p>And in order for software to be reliable, its codebase must be understandable by the humans who are building and maintaining it. A probabilistic algo isn’t likely going to succeed in automating that away (as <a href="https://visualstudiomagazine.com/articles/2024/01/25/copilot-research.aspx">some engineers are starting to discover this year</a>).</p><p>Intention is better than an elaborate attempt to reverse engineer the odds of intention. I expect Code Connect will solve this problem better than AI can.</p><p>I think it’s worth trying out Code Connect, if you have access to the right Figma license (cautiously, as it’s in beta, and subject to change).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0b7eaf7a068" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The future of design systems is no design systems]]></title>
            <link>https://medium.com/@majtom2grndctrl/the-future-of-design-systems-is-no-design-systems-c4baa6b64949?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/c4baa6b64949</guid>
            <category><![CDATA[front-end-engineering]]></category>
            <category><![CDATA[ui-ux-design]]></category>
            <category><![CDATA[april-fools]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Mon, 01 Apr 2024 16:17:09 GMT</pubDate>
            <atom:updated>2024-04-02T19:03:33.932Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VXxy-0Wv597HsVmQ5OoMnw.jpeg" /></figure><h4>This was an <strong>April Fool’s Day joke</strong>. It’s a short read, so if you’re up for a cathartic chuckle, read on!</h4><p>Design systems were a really big deal in the latter half of the 2010s. But like all new technologies, they eventually became outdated. Modern product teams have moved beyond design systems, to bold new methodologies!</p><h3>Research Revolution? More Like Research… Shmearch!</h3><p>Design systems promised a utopia where teams focused on solving user problems, not button placement. But we found that this efficiency eliminated any available time slots previously reserved for user research!</p><p>To better understand why our teams didn’t have time for user research, our amazing team of product managers did their own research (read: late-night brainstorming session fueled by questionable snacks) and discovered that users <strong>thrive</strong> on inconsistency. Humans were evolved from hunter-gatherers — finding those difficult-to-find buttons is what we’re born for!</p><h3><strong>Engineering Unleashed!</strong></h3><p>Design system governance structures were an unnecessary hassle. Engineering teams crave autonomy, and what’s more autonomous than coding interfaces that look like a Jackson Pollock painting come to life?</p><p>Once we liberated our engineering teams to do what they want, we also found that generative AI was much better at building UIs that look like Jackson Pollock paintings than humans—and faster—unlocking next-generation design efficiency.</p><p>So ditch the design system rulebooks and embrace the beautiful mess that is a user interface designed by a team unleashed. Let generative AI be your wild paintbrush, and celebrate the inherent human desire to click on a misplaced button that might (or might not) lead them to the checkout page.</p><p>The future of UI/UX design has never been so bright!</p><p>(If it wasn’t clear, this article is satire. It was also drafted with the help of generative AI, and some human-edits. Happy Monday!)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c4baa6b64949" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Creating a color scale with AI, or: How to co-hallucinate with AI to prototype an abstract idea]]></title>
            <link>https://medium.com/@majtom2grndctrl/creating-a-color-scale-with-ai-or-how-to-co-hallucinate-with-ai-to-prototype-an-abstract-idea-eb7f6777b111?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/eb7f6777b111</guid>
            <category><![CDATA[ui-ux-design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[product-design]]></category>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Wed, 17 Jan 2024 21:41:17 GMT</pubDate>
            <atom:updated>2024-01-18T20:03:27.583Z</atom:updated>
            <content:encoded><![CDATA[<p>Over the holiday break, my portfolio site got to the point where I needed to ensure color harmony by bringing in a color scale. When I started looking around at color scales, though, I realized I didn’t want to do my color scale with colors that blend like a linear gradient.</p><p>Early in my career, I noticed that if I boost the saturation on the very dark and very light colors, that helps make the hue more perceivable, and connects those colors to the brand color more effectively.</p><p>Keeping this in mind, I had an idea for a “dream” color scale design tool. It might look something like this:</p><figure><img alt="An image of an HSL color selector where a blue hue has been chosen, with a bezier curve layered over the top that has nine points." src="https://cdn-images-1.medium.com/max/1024/1*OaPObgbm6VUZT9JWBOKiyA.png" /><figcaption>A rough mockup of what my dream color scale tool might look like.</figcaption></figure><p>I wanted to draw a bezier curve that would allow me to quickly choose a color scale. The white dots represent colors that I chose manually (the darkest, middle, and lightest colors), and the gray dots represent the colors in between, chosen for me by the tool.</p><p>There’s just one problem: I couldn’t find a color scale tool that works like this. (If you know one, I’d love to hear about it!)</p><p>So, instead of trying to choose all of these colors by hand, I had an idea: <em>What if I </em><strong><em>prototype the concept</em></strong><em> for my color scale using an LLM?</em></p><p>At worst, I’ll have a funny story to tell. At best, I’ll get the AI to co-hallucinate a color scale with me!</p><h3>First draft of my ChatGPT prompt</h3><p>Asking an AI to create a color scale — here’s how it started:</p><blockquote>Help me create a color scale based around the color #38A0CD, using HSL color. I want there to be nine steps to the scale. I want the nine steps to use HSL color, and form a parabolic curve, where the lighter and darker colors have more saturation. Let’s call the color brand blue. Output this color scale as a series of CSS custom properties.</blockquote><p>In the first draft of the scale it gave me, I wasn’t satisfied with the colors ChatGPT chose as the start and end points of the scale, so I asked it a follow-up question:</p><blockquote>Let’s re-do the parabolic curve a little. Can we make the dark point of the curve hsl(198, 97, 7), the middle point of the curve hsl(198, 60, 51), and the light point of the curve hsl(198, 95, 95)?</blockquote><p>From this, I got:</p><pre>  --brand-blue-1: hsl(198, 97%, 7%);    /* Darkest */<br>  --brand-blue-2: hsl(198, 88%, 25%);<br>  --brand-blue-3: hsl(198, 78%, 34%);<br>  --brand-blue-4: hsl(198, 69%, 43%);<br>  --brand-blue-5: hsl(198, 60%, 51%);   /* Middle */<br>  --brand-blue-6: hsl(198, 73%, 68%);<br>  --brand-blue-7: hsl(198, 80%, 78%);<br>  --brand-blue-8: hsl(198, 87%, 87%);<br>  --brand-blue-9: hsl(198, 95%, 95%);   /* Lightest */</pre><figure><img alt="Two sets of nine color samples. One was sampled from a gradient, and one uses colors from ChatGPT. The ChatGPT colors are slightly easier to recognize as being different from each other." src="https://cdn-images-1.medium.com/max/1024/1*PPqQ0sZixrwmhKH1I5e2tA.png" /></figure><p>For the purposes of the idea I wanted to test, this did what I wanted. The difference between each shade is easier to notice, and this can help each shade stand out better when placed close to another similarly colored object.</p><p>This could have been the end of the story. But as I found out through repeated experimentation, this was an <em>extraordinary</em> case of beginner’s luck.</p><h3>If at first you <em>think</em> you succeed, try, try again 😂</h3><p>When I evaluate a new technology or technique for work, it often isn’t enough to see that it works once. I need to understand <em>how</em> it works. Or at least, I need to see it work more than once.</p><p>Also, I’m working with OKLCH colors on my project, so my next step was to ask the LLM to generate the scale with OKLCH a few more times, with a slightly different prompt each time, to see what I get.</p><p><em>This is where it started to get weird.</em></p><figure><img alt="A set of 11 color swatches, ranging from dark to light. Near the lighter end of the scale, the colors become very saturated and overly bright, such that they don’t blend into the last step in the color scale." src="https://cdn-images-1.medium.com/max/1024/1*qNc8f6gaKq9VVrkUeOgCjw.png" /><figcaption>What’s wrong with this color scale? 🤣</figcaption></figure><p>When I asked the LLMs to set the scale in OKLCH colors, I kept getting colors at the lighter end of the scale that didn’t even come close to blending in to the final step of the scale. I tried dozens of times to reframe my prompt in such a way that I could get a different result. I tried ChatGPT 3.5 via Open AI, ChatGPT 4 via Microsoft Copilot, and even Google Bard. But every color set came back looking like this.</p><p>Along the way, there were a handful of other responses that were not what I wanted.</p><p>For example, this code sample used variables that it didn’t declare, so none of these colors worked:</p><pre>    --brand-blue-darkest: oklch(20.2% 0.048 223);<br>    --brand-blue-lightest: oklch(96.4% 0.027 221.6);<br>    --brand-blue-middle: oklch(66.1% 0.14 224.6);<br>  <br>    --brand-blue-step-1: oklch(<br>      calc(20.2% + (var(--brand-blue-middle-luminance) - var(--brand-blue-darkest-luminance)) / 5) <br>      calc(0.048 + (var(--brand-blue-middle-saturation) - var(--brand-blue-darkest-saturation)) / 5) <br>      calc(223 + (var(--brand-blue-middle-hue) - var(--brand-blue-darkest-hue)) / 5)<br>    );</pre><p>This sample used the same color for the last three steps. Also, on step 8, it set the luminance value above 100%. OKLCH doesn’t support that.</p><pre>  --brand-blue-1: oklch(20.2% 0.048 223); /* Darkest */<br>  --brand-blue-2: oklch(31.6% 0.077 224.2);<br>  --brand-blue-3: oklch(42.9% 0.109 224.4);<br>  --brand-blue-4: oklch(54.3% 0.134 224.5);<br>  --brand-blue-5: oklch(66.1% 0.14 224.6); /* Middle */<br>  --brand-blue-6: oklch(77.9% 0.134 224.5);<br>  --brand-blue-7: oklch(89.3% 0.109 224.4);<br>  --brand-blue-8: oklch(100.6% 0.077 224.2);<br>  --brand-blue-9: oklch(96.4% 0.027 221.6); /* Lightest */<br>  --brand-blue-10: oklch(96.4% 0.027 221.6);<br>  --brand-blue-11: oklch(96.4% 0.027 221.6);</pre><p>Also, I found that converting colors from HSL to OKLCH is not straightforward. In Figma, I use a plugin called <a href="https://www.figma.com/community/plugin/1173638098109123591/okcolor">OkColor</a>, which itself uses a JavaScript library called <a href="https://culorijs.org/">Colouri</a>. The LLMs apparently haven’t learned this library’s secrets, as of holiday season 2023. In general, I found LLMs don’t generate good color scales in OKLCH.</p><p>Finally, I realized a parabolic curve was too rigid or literal of a concept to ask an LLM for. Moving forward, I ask for a <strong>bezier curve</strong> instead.</p><h3>The results, after so much tinkering</h3><p>Out of all the colors scales I generated, no two scales were the same. And even though I kept trying to improve the prompt, none of the scales I tested generated a scale of colors I liked better than the first one.</p><p><strong>Pro tip:</strong> In addition to having the LLM generate color scales for you, you can also have it generate code with color swatches that demonstrate the colors, so you can rapidly see them next to each other.</p><h3>What did I learn, here?</h3><p>First of all, I was reminded that <strong>LLMs are not perfect</strong>. I could use the exact same prompt several times and get different color values from each attempt. ChatGPT 3.5, especially had a tendency to tell me it did something that it didn’t actually do—but I had the same problem to lesser degree with the other LLMs, too.</p><p>Also, I learned that <strong>LLMs don’t do math</strong> as computers normally do. If an LLM has been trained on a lot of math stuff, it memorizes the calculations and hallucinates math stuff based on what it learned. (Which is pretty relatable, isn’t it?!)</p><p><strong>For now, LLMs can do a first draft faster than us</strong>, but the first draft is never perfect. Learning to work well with an LLM means figuring out how to set up the LLM to set us up for success. It frequently won’t do what you want on the first try, so it helps to learn how to ask it for revisions.</p><p>But my big takeaway here is: Don’t just ask an LLM to give you a thing. <strong>Bring your personal spark to the prompt. </strong>For me, that spark was asking for the color curve. You might have your own, different strategy for setting up a color scale. One strategy may work well for one project, and another strategy may work well for a different project. As a human, it’s still up to you to own the vision!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eb7f6777b111" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Googled “font ramp,” and it opened my eyes to so much more]]></title>
            <link>https://medium.com/@majtom2grndctrl/i-googled-font-ramp-and-it-opened-my-eyes-to-so-much-more-e19d5b404426?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/e19d5b404426</guid>
            <category><![CDATA[token-design]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[ui]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Mon, 11 Dec 2023 00:15:27 GMT</pubDate>
            <atom:updated>2023-12-12T05:47:40.413Z</atom:updated>
            <content:encoded><![CDATA[<p>What I found made me “work smarter, not harder” with little details that make a big difference.</p><p><strong>Note:</strong> This article was originally published in 2019 on my website. It’s been re-published here, with some new thoughts.</p><p>I’ve heard the term, “font ramp,” at the office, and I always assumed the meaning of it was obvious, from looking at it. But it wasn’t mentioned in my design training, so I <em>didn’t actually </em><strong><em>know</em></strong><em> what it meant.</em></p><p>In May 2019, I Googled it, did a <a href="https://type-scale.com/">deep dive</a> <a href="https://www.modularscale.com/">on</a> <a href="https://24ways.org/2011/composing-the-new-canon">type scales</a> — and felt an irresistible urge to refactor the CSS on my website to use type scales for everything! <em>What could possibly go wrong with a site-wide CSS refactor?</em> 😅</p><h3>Getting started with a type scale</h3><p>I started by creating two variables:</p><ol><li><strong>— type-scale-base</strong>, the base font size of my main body copy.</li><li><strong>— scale-multiplier</strong>, the number I want to use to scale from one font size to the next.</li></ol><p>From there, I added the type-scale variables, generated from combinations of the two variables above.</p><pre>--type-scale-1: var(--type-scale-base);<br>--type-scale-2: calc(var(--type-scale-1) * var(--scale-multiplier));<br>--type-scale-3: calc(var(--type-scale-2) * var(--scale-multiplier));<br>--type-scale-0: calc(var(--type-scale-1) / var(--scale-multiplier));<br>--type-scale-00: calc(var(--type-scale-0) / var(--scale-multiplier));</pre><p>And so on. My full range of variables goes from <strong>— type-scale-00</strong> to <strong>— type-scale-8</strong>.</p><h3>Next, I applied the type scale to everything. No exceptions.</h3><p>It’s natural to worry that applying a predefined type scale to everything can be too rigid, but I found a few helpful tricks for this along the way.</p><p><strong>Trick #1: Experiment with changing the multiplier.</strong> When the type scale felt like it was too rigid, I think the <em>real</em> problem was that I chose the wrong multiplier.</p><figure><img alt="Three screenshots of the same homepage design, one where spacing is large, one where spacing is tight, and one in between." src="https://cdn-images-1.medium.com/max/1024/1*5LzJqEuqE_agPaORI8qYEQ.jpeg" /></figure><p><strong>Same layout, different scale multiplier. </strong>Trying out different multipliers helped me see multiple options quickly. With a higher multiplier, big fonts are comparatively bigger; with a lower multiplier, big fonts are still bigger, but less big. Above left: 1.333x, or a “perfect fourth” scale. Above right: 1.2x, or a “minor third” scale. Below: 1.25x, or a “major third” scale. The “major third” scale felt “just right” for desktop form factors.</p><p><strong>Trick #2: Use media queries to change CSS variables.</strong> For example, if the font scale feels great on a desktop device, but feels like too many things get too big on mobile, <em>you can use a smaller multiplier on smaller devices</em>.</p><p>I ended up going with a multiplier of <strong>1.2</strong> (a “minor third” scale) on mobile devices, and a multiplier of <strong>1.25</strong> (a “major third” scale) on desktop screens.</p><h3>The type scale alone wasn’t enough, so I added a “spatial scale” to the mix.</h3><p>One of the biggest mistakes I often see UI/UX designers make with typography is to assign a line-height in CSS according to a “best practice” document they read.</p><p>Most of what we read lives in the space between the baseline and the cap-height of each font. You know how a line-height of 200% looks like it has one line of whitespace between each line of text for some fonts, but it looks like it has 1–½ lines of whitespace for other fonts? That’s because different fonts have different cap-heights.</p><p>This variability is a hard constraint of working with fonts, so to compensate, I created a set of new variables in my CSS, called <strong>spatial scale</strong>. To find the spatial scale base, I reduced the line-height on a large heading until the element’s borders got as close as possible to lining up with the baseline and cap-height.</p><figure><img alt="A screenshot showing a web inspector being used to find how tall the capital letters are in the Poppins webfont." src="https://cdn-images-1.medium.com/max/1024/1*6SJo0SPwHZymIvFvuoGvZA.jpeg" /></figure><p><strong>Here’s what the cap height looks like for Poppins.</strong> I adjusted the line-height of a heading until its bounding box was as close to lined up with the cap height as possible.</p><p>Next, I setup variables like these:</p><pre>--spatial-scale-base: calc(var(--type-scale-base) * .705);<br>--spatial-scale-1: var(--spatial-scale-base);<br>--spatial-scale-2: calc(var(--spatial-scale-1) * var(--scale-multiplier);<br>--spatial-scale-3: calc(var(--spatial-scale-2) * var(--scale-multiplier))</pre><p>I find it helps for the spatial scale to be larger than type scale, so I have a range from from <strong>— spatial-scale-00</strong> to <strong>— spatial-scale-14</strong>.</p><p>I used this spatial scale for almost all <strong>line-height</strong>, <strong>margin</strong>, <strong>padding </strong>rules. I also have a rule that sets --grid-gutter: var(--spatial-scale-2);.</p><h3>Before and after impression</h3><p>The visual differences came out a lot smaller than you might think, from <a href="https://github.com/majtom2grndctrl/distantlyyours-v8/pull/20/files">looking at how much code was changed</a>.</p><p>In some places, I felt like this exercises validated that my original eye for composition had matured more over the years than I had given myself credit for.</p><p>In others, this system held me accountable for the places where I ran out of time or energy and started to just put things in place quickly. And better than that: I had a tool for implementing intentionally proportioned design quickly.</p><figure><img alt="Two screenshots showing my site, before and after this exercise." src="https://cdn-images-1.medium.com/max/1024/1*Bsn4g8yPLkNR27sh9cu_qA.jpeg" /></figure><p><strong>Before and after:</strong> The page title, where I had been very deliberate in my first pass about sizes and spatial relationships. I intentionally had equal whitespace above and below the page title the first time, but having spatial scale variables helped me try out other precisely calculated proportional relationships quickly.</p><figure><img alt="A screenshot of the top section of my site, with some annotations showing the specific heights of each part of the layout, to visually show how the spatial relationships work." src="https://cdn-images-1.medium.com/max/1024/1*yUb0hRu5x95O6iLiKVQIsg.jpeg" /></figure><p><strong>After:</strong> Here, the spatial scale is overlaid on top of the layout. You can see the whitespace between the logo and the section title is equal to the cap-height of the section title. The whitespace between the section title and the content is equal to the height of the logo. The logo itself is set to a height from the spatial scale, which means the logo’s height is connected to the base font size.</p><figure><img alt="Two screenshots showing what typography on my site looked like before and after this exercise." src="https://cdn-images-1.medium.com/max/1024/1*2hRzPdBFj1ExI8DUQut0dA.jpeg" /></figure><p><strong>Before and after:</strong> CMS content. Before, I had set all margins in em units, which meant whitespace between each block changed with the font size. Spatial scale variables made it painless to set one margin size derived from the base font size. If the font size changes at a later date, the margins automatically change proportionally.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QWjDbR50kSNaQ5Ecj4cxzQ.jpeg" /></figure><p><strong>Before and after:</strong> Portfolio case study heading. Before, I had made guesses about line height for the large text, and whitespace between elements. Using variables derived from the base font size helped me iterate with confidence as I adjusted what size the pieces were and how they fit together.</p><h3>Lessons learned from mistakes along the way</h3><p><strong>Always start with the base content font size.</strong> Most of my design time goes to layouts that don’t have content. Because this started out as a passion project, I started out with defining font sizes and scales for the portfolio page. But when I moved deeper into case studies and blog posts, I found my base font sizes didn’t work for plain text content. Oops.</p><p><strong>The first refactor is the hardest, but it gets easier.</strong> The first refactor took about a week of spare time. Unfortunately, I noticed mistakes at the end, and had to refactor again. The second and third refactors each took about ⅙ as much time. This was partially because the items I needed to change were fresh on my mind but partially because having a predefined system of sizes nearly eliminated analysis paralysis.</p><h3>What I like about this approach</h3><p>I feel like this method brings a greater sense of order and visual consistency to the layout, without making the layout feel too rigid. It feels more organic than something that was laid out on graph paper.</p><p>In a lot of other baseline grid systems I’ve seen, like a grid where every font size, margin, padding, and grid gutter is set to a multiple of 4px, I think it’s hard to see a sense of order when some spaces are 40px and the next space is 44px. Setting these measurements as exponential values rather than linear incremental measurements removes extraneous options — increasing the sense of visual order, and limiting analysis paralysis.</p><h3>Challenges to address next</h3><p>This process also poses some challenges to designers who only use Figma. My UI code manages font sizes, line heights, and grid gutters down to fractions of pixels. It’s hard work to make Figma match that level of fidelity. Using Figma variables and design tokens can help with at least communicating the name of each measurement.</p><p>That said, let’s talk about how design jobs are designed for a moment.</p><p>Because I have the ability to both design and build user interfaces and websites, I <em>know</em> it’s very inefficient to expect designers to toil over pixel-perfect mockups, so that engineers can toil over pixel-perfect replicas of the design in code. Although design tools have helped designers move faster by helping them build rules-based systems directly in the design docs, they still have a long way to go to match the power of CSS itself.</p><p>This is a great example of the kind of problems design systems help solve. It also shows why a true design system is made up of deliverables from a cross-functional group of contributors. With a less siloed design process, the product team can spend less time on finicky visual design problems, and spend more time on problems that have deeper consequences for the business and its end users.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e19d5b404426" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Don’t be shocked by Brutalist Web Design, check for a rationale]]></title>
            <link>https://medium.com/@majtom2grndctrl/dont-be-shocked-by-brutalist-web-design-check-for-a-rationale-aed5c7101949?source=rss-59bf9e97ac95------2</link>
            <guid isPermaLink="false">https://medium.com/p/aed5c7101949</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[design-principles]]></category>
            <dc:creator><![CDATA[Dan Hiester]]></dc:creator>
            <pubDate>Thu, 12 May 2016 10:42:29 GMT</pubDate>
            <atom:updated>2016-05-12T10:49:13.976Z</atom:updated>
            <cc:license>http://creativecommons.org/licenses/by/4.0/</cc:license>
            <content:encoded><![CDATA[<p>If you grew up in the ’90s, you probably remember a lot of edgy graphic design. Back then, the Wipeout video game series was a gateway drug that introduced me to The Designers Republic. I remember the Future Sound of London’s site being one of the earliest rich media experiences I saw on the web, an angsty and foreboding experience that was just as much an expression of angst and social disconnection as the group’s seminal 1996 album, Dead Cities.</p><p>When Radiohead released A Moon Shaped Pool to the public, <a href="http://www.amoonshapedpool.com/">the album’s website</a> filled me with nostolgia. There it was, in all caps: “RADIOHEAD”. In every variation of Radiohead’s logo, I always felt like it had a particularly hideous capital A. This new site was the real deal. Back in the day, in order to get such ugly, unsettling-looking typography onto a website, we had to resort to using a gif, or a Flash movie. But now, in 2016, we can reproduce that angsty look with embeddable web type. Hooray for progress!</p><h3>A web design gallery recieves mainstream media coverage for being so ugly</h3><p>When I read the Washington Post’s <a href="https://www.washingtonpost.com/news/the-intersect/wp/2016/05/09/the-hottest-trend-in-web-design-is-intentionally-ugly-unusable-sites/">recent article on “Brutalist Web Design”</a>, however, the nostalgia train came to a halt. The piece, covering <a href="http://brutalistwebsites.com/">brutalistwebsites.com</a>, drew obvious comparisons with the popular architectural style of the 1970s, and pointed out that some brutalist-looking sites have always been popular. Reading through the piece, and some of the comments, I was surprised how many people acted shocked that anyone would <em>intentionally</em> make something ugly.</p><p>“This is web art not web design”, one commenter said.</p><p>“Congratulations on making your philosophical point, you design rebels, you”, said another.</p><p>These two comments make some valid observations, but I think under a lens of constructive criticism, a designer can take these points further. I think a designer <em>should</em> look at these sites constructively. Pointing to a problem without offering a solution is also known as whining. A designer’s job isn’t to whine about bad design, our job is <em>to make it better</em>.</p><p>I’m sharing this because I’ve participated in some conversations in the past year that left me with the impression that a lot of people don’t keep this in mind, when they talk about design.</p><h3>To connect to an emotion, a design must feel authentic to its audience</h3><p>The first comment I quoted makes a fair point. If all creative work rests somewhere in a continuum between form and function, these brutalists sites lean more in favor of form than function. Having said that, people who work with design every day shouldn’t be shocked at an intentionally unattractive design. There are plenty of valid reasons, which often <em>aren’t strictly philosophical</em>.</p><p>One example: Brands need to feel authentic to connect with audiences.</p><p>While I haven’t seen any brand guidelines for Radiohead, I feel like I can guess some of the themes they want to evoke. Anyone should be able to look at a Radiohead album cover, and be able to tell it isn’t music popular people like. Radiohead’s music connects to people who’ve <em>never</em> felt popular. Or maybe it’s the <em>one thing</em> that connects with the popular people when they have a bad day. But in order to discover the music, it helps them to <em>see in </em><strong><em>advance</em></strong> how edgy and distraught the music will be.</p><p>An album full of upbeat party tunes should come with album artwork conveying <em>that</em> feeling, too. If the music is so upbeat that it feels manufactured and insincere, the graphic designer can let those words guide their work.</p><p>The process of choosing words to guide the direction of a project is often covered by design principles.</p><h3>Design principles clarify the otherwise murky waters of a project’s creative objectives</h3><p>As designers, our job is to understand <em>which visuals</em> evoke <em>which emotions</em>. It’s helpful to make sure a project’s stakeholders agree about <em>which</em> emotions they want to convey, and share some consensus of <em>how</em>. Mood boards can help you ask questions, like, “What do you like about this particular design? What feeling do you get from looking at it, and why do you think it affects you that way?” After everyone gets their cards on the table, you are in a much better position to steer the project in the best direction.</p><p>Without having that conversation, different stakeholders often have different visions for a project, resulting in <em>them</em> sending <em>you</em> mixed signals. This gives you a false-start, and wastes precious budgeted time.</p><p><em>Design is subjective, </em><strong><em>in a vacuum</em></strong>. By establishing an agreed-upon context for creative decision making, you can help remove a layer of subjectivity. <strong>Do not create an edgy looking design because you think it looks cool.</strong> That drags your project back down into subjectivity.</p><h3>Design principles aren’t just for stakeholders, they’re there for you, too</h3><p>Even if your stakeholders couldn’t be bothered with design principles, at least set them up <em>for you</em>. Ask yourself questions, like, “No, seriously, <em>what emotion </em><strong><em>really</em></strong> speaks to the heart of this project?” Give yourself some space to develop a theory about why you want to evoke that emotion. Go nuts, create a <a href="http://alistapart.com/article/crafting-a-design-persona">design persona</a>.</p><p>I’m obsessed with music, so I ask myself questions like, “If this site had its own personality, what music would it like? What songs connect with it differently than with me? Why is that?”</p><p>After pondering these questions, foundational steps like choosing typefaces, or defining a color palette, will have a greater sense of purpose, and intent. I have one less voice in my head, saying, “Well, maybe you should try <em>one more</em> thing?”</p><p>It gets me one critical step closer to <strong>done</strong>.</p><p>This post was <a href="http://distantlyyours.com/blog/dont-be-shocked-by-a-brutalist-web-design-check-for-a-rationale">originally published</a> at my personal site, <a href="http://distantlyyours.com/">Distantly Yours</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aed5c7101949" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>