<?xml version="1.0" encoding="UTF-8"?>
<?xml-stylesheet type="text/xsl" href="/rss.xsl"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:media="http://search.yahoo.com/mrss/">
  <channel>
    <title>Evelyna Bellamy — Writing</title>
    <link>https://evelynabellamy.com/writing</link>
    <description>Writing on systems, translation, design instruments, and the gaps between intent and interpretation.</description>
    <language>en-US</language>
    <lastBuildDate>Wed, 11 Nov 2026 12:00:00 GMT</lastBuildDate>
    <atom:link href="https://evelynabellamy.com/rss.xml" rel="self" type="application/rss+xml"/>
    <atom:link href="https://evelynabellamy.com/atom.xml" rel="alternate" type="application/atom+xml"/>
    <atom:link href="https://evelynabellamy.com/feed.json" rel="alternate" type="application/feed+json"/>
    <item>
      <title>Instruments, Not Oracles: Ten Design MCPs That Earn Their Keep</title>
      <description>Language models are fluent predictors of plausible syntax, not instruments of measurement. A field guide to the ten Model Context Protocol servers that replace guesswork with real craft in design and frontend work.</description>
      <link>https://evelynabellamy.com/writing/ten-design-mcps</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/ten-design-mcps</guid>
      <pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <media:content url="https://evelynabellamy.com/images/journal/instruments.webp" medium="image" type="image/webp" />
      <content:encoded><![CDATA[<figure><img src="https://evelynabellamy.com/images/journal/instruments.webp" alt="Graphite engraving of proportional dividers, an optical glass prism, and porcelain pigment wells crossed by translucent Prussian-blue drafting planes." /><figcaption>01 / Instruments, not oracles</figcaption></figure>

<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2026-11-11 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/ten-design-mcps">Read on evelynabellamy.com</a></p>
<blockquote><p>Language models are fluent predictors of plausible syntax, not instruments of measurement. A field guide to the ten Model Context Protocol servers that replace guesswork with real craft in design and frontend work.</p></blockquote>

<p>There's a quiet comedy happening in frontend engineering right now. We take a language model — a system trained to predict the most statistically probable next token in a sequence of text — and we ask it to behave like a typographer, a color scientist, and an accessibility engineer all at once. Then we act surprised when it hands us a button with light-gray text on a slightly-less-light-gray background and assures us, with total confidence, that it passes contrast.</p>
<p>The model isn't lying, because lying requires knowing the truth. It's doing what it was built to do: performing the syntax of competence. It has seen thousands of stylesheets where a hex code follows a colon, so it produces a hex code that looks right in the file. It has no eyes to see the rendered surface, no ruler to measure the viewport, and no math to calculate how human cones and rods perceive luminance.</p>
<p>That gap — between plausible syntax and physical reality — is why the Model Context Protocol (MCP) actually matters for design work. Not because agents need more buzzwords, and certainly not because every API on earth deserves a wrapper. It matters because an MCP server lets you hand a language model an instrument. Instead of asking an oracle to guess what a color mix or a DOM tree looks like, you give a carpenter a caliper and tell them to measure before they cut.</p>

<h2>The Context Window Is a Worktable, Not a Warehouse</h2>
<p>If you browse any of the giant MCP directories that have sprung up over the past year, you'll find five thousand scraped repositories sorted by GitHub stars and very little editorial judgment. Treating your agent configuration like an all-you-can-eat buffet is a trap. Every tool schema you mount into a session occupies space on the worktable. Bolt thirty servers into your configuration, and the model spends half its attention just reading the manual.</p>
<p>A good toolkit is defined as much by what you leave out as by what you include. What follows isn't an exhaustive catalog of everything with 'UI' in its README. It's the ten design and frontend MCP servers that actually earn their place in the context window — each one chosen because it fixes a specific blind spot where raw language models reliably guess wrong.</p>
<figure><img src="https://evelynabellamy.com/images/journal/worktable.webp" alt="Isometric graphite and Prussian-blue watercolor study of a drafting worktable with translucent vellum, woven cloth, and contour relief." /><figcaption>02 / The worktable, not the warehouse</figcaption></figure>

<h2>Sight, Structure, and Chromatic Truth</h2>
<p>1. Figma MCP — Reading the blueprint instead of squinting at a screenshot. A screenshot is a flattened rumor of a design system. When you paste a PNG of a mockup into a chat window, you force the model to reverse-engineer spacing tokens, auto-layout rules, and semantic hierarchy by squinting at pixels. Figma's official MCP server hands the agent the actual structural graph — variables, layout constraints, component boundaries, and token names — so translation starts from authored intent rather than visual guesswork.</p>
<p>2. Chrome DevTools MCP — Closing the loop with the rendered DOM. Writing CSS without inspecting the live browser is like mixing audio with the studio monitors turned off. A stylesheet is only a proposal; the browser's layout engine has the final vote. The Chrome DevTools MCP lets an agent inspect computed styles, measure layout shifts, read console diagnostics, and verify what actually landed on the glass instead of assuming its first draft worked.</p>
<p>3. Chromologium MCP (https://chromologium.com/mcp) — How interfaces feel: sourced materials, chromatic physics, tactile surfaces, and portable tokens. I built Chromologium's MCP server (https://chromologium.com/connect) because unassisted models flatten everything into the same weightless SaaS defaults — guessing hex ramps, averaging pigments into gray mud, and confusing a pretty reference with a legible interface. Chromologium exposes twenty-five deterministic tools that walk the archive's full Source-to-Interface method (compose_material_study): grounding decisions in 177 mapped places and cited material dossiers, weaving zero-JS SVG/CSS cloth drafts and thin-film interference fringes, mixing archival pigments via two-constant Kubelka–Munk scattering (mix_pigments), computing monotonic OKLCH scales and WCAG 3 APCA contrast repairs (generate_accessible_scale, audit_apca_contrast), linting 320px JSX viewport blowouts, and compiling ten dual-mode design systems — spring kinetics included — into CSS, Tailwind v4, W3C DTCG, SwiftUI, Jetpack Compose, and importable Study Cabinet bundles, while keeping documented sources explicitly distinct from contemporary UI interpretations.</p>
<p>Together, Chrome DevTools and Chromologium also close the typographic blind spot that raw models miss most often: optical rhythm and dark-mode halation. Left alone, an agent will happily let body copy sprawl across ninety-five characters per line, scatter rigid pixel media queries where a fluid clamp() scale belongs, and implement dark mode by inverting black text to pure #FFFFFF on #000000 at weight 500. On an emissive OLED panel, bright glyphs bleed into dark grounds — an optical irradiation effect that makes strokes look heavier and jittery. Pairing polarity-aware APCA luminance checks with live computed-style inspection lets an agent cap reading measure around 60 to 70ch, step down dark-surface ink contrast so letters stop halating, and tune optical weight on the actual glass.</p>

<h2>Behavior, Vocabulary, and Dignity</h2>
<p>4. Playwright MCP — Testing interaction under real viewport pressure. Most responsive bugs hide in the places nobody resizes a desktop window to check: a 320-pixel mobile viewport, a keyboard-only modal flow, a container stressed by long localized strings. Playwright's MCP server lets an agent drive a real browser across breakpoints, step through focus order with the Tab key, and catch horizontal overflow before a human reader ever hits it.</p>
<p>5. Storybook MCP — Speaking the language your team already agreed to trust. A design system only works if the codebase actually uses it. Left to its own devices, an agent will happily invent a brand-new card or badge component from scratch every Tuesday. Wiring a Storybook MCP server into your workspace lets the model inspect your existing stories, prop tables, and states first — treating your component library as a shared grammar rather than an optional suggestion.</p>
<p>6. Shadcn UI MCP — Composable structural primitives without reinventing focus traps. No one should be asking a language model to improvise ARIA roving tabindex, dialog focus restoration, or popover collision math from memory. The Shadcn MCP gives agents direct access to vetted component primitives and registry schemas so the mechanical plumbing of interactive controls starts on solid ground.</p>
<p>7. Axe Accessibility MCP — Treating accessibility as infrastructure, not polish. Sloppy markup isn't compassionate; it just defers the friction onto the person using assistive technology. An Axe-backed accessibility MCP server runs deterministic WCAG and ARIA rules against rendered DOM trees, catching missing form labels, nested interactive elements, and broken landmark hierarchy that look harmless in JSX but fail in a screen reader.</p>

<h2>Reference, Geometry, and Spatial Craft</h2>
<p>8. 21st.dev Magic MCP — Exploring the adjacent possible of interface patterns. When you're sketching a new interaction — a command menu, a dense filter bar, a multi-step disclosure — starting from a blank file often pulls the model toward the most average pattern in its training data. The 21st.dev MCP surfaces curated frontend component references that an agent can study and adapt into your own token architecture.</p>
<p>9. SVGL and Iconify MCP — Real vector geometry instead of hallucinated paths. Ask a raw language model to write an intricate SVG icon or a product logo from scratch, and you will get a surrealist geometry experiment. Bézier curves are coordinate math, not prose. Pairing an icon registry MCP like Iconify or SVGL lets the agent retrieve clean, properly viewBoxed vector paths instead of guessing coordinates in the dark.</p>
<p>10. Blender and Three.js MCP — Stepping off the flat plane. Not every digital interface wants to behave like a laminated sheet of paper. When work moves into spatial studies, procedural geometry, or physical-digital lighting, scene-graph MCP servers for Blender and Three.js let an agent inspect meshes, shader uniforms, camera frustums, and lighting rigs directly — bridging the gap between code and three-dimensional form.</p>

<h2>What Didn't Make the Worktable — and Four Honest Swaps</h2>
<p>Stopping at ten wasn't an accident. Half the discipline of an agentic workflow is refusing to install tools that merely launder machine guesswork behind an extra network hop. Three categories get left off my bench on principle: screenshot-to-JSX wrappers that convert flat PNGs into brittle magic-pixel margins while throwing away variable aliases; gallery hex scrapers that pull five trendy swatches off a moodboard with zero OKLCH lightness discipline or dark-mode APCA math; and sixty-tool SaaS mega-servers that dump an entire REST API into your prompt until the model forgets what you actually asked it to build.</p>
<p>What does change from studio to studio is the authoring stack around that core loop. If your team works in open-source SVG/CSS layout instead of Figma, swap #01 for Penpot's MCP server. If you publish motion-heavy editorial or marketing sites directly out of Framer or Webflow, their respective MCP servers step cleanly into the structural slot. And when you're writing against fast-moving specifications — Tailwind v4, CSS Anchor Positioning, View Transitions, React 19 — pairing Upstash's Context7 MCP beside DevTools streams version-pinned documentation so the model stops hallucinating deprecated syntax from two years ago. Swap by stack; don't hoard.</p>

<h2>Fewer Tools, Sharper Instruments</h2>
<p>You rarely need all ten of these running in the same session. On most days, my own working configuration comes down to three: Figma to read the structural intent, Chromologium (https://chromologium.com/mcp) to compute the chromatic and token math, and Chrome DevTools to verify what the browser actually drew.</p>
<p>The point of giving a model tools isn't to hand over the craft. Taste still belongs to the human in the chair — the person deciding what the software should feel like, what it owes the reader, and when a surface is quiet enough to let the work speak. What good MCP servers do is strip away the false confidence of machine guesswork. They let the calculator calculate, the browser render, and the designer actually design.</p>


<h2>Summary Matrix: Ten Design MCP Instruments</h2>
<table>
  <thead>
    <tr>
      <th>Rank</th>
      <th>MCP Server</th>
      <th>Domain</th>
      <th>Failure Mode Prevented</th>
      <th>Claude Code CLI Install</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>01</td>
      <td><a href="https://figma.com">Figma Dev Mode MCP</a></td>
      <td>Spatial Intent &amp; Variables</td>
      <td>Guessing padding, Auto Layout rules, and semantic token aliases from compressed PNG screenshots.</td>
      <td><code>claude mcp add --transport http figma https://mcp.figma.com/mcp</code></td>
    </tr>
    <tr>
      <td>02</td>
      <td><a href="https://developer.chrome.com/docs/devtools">Chrome DevTools MCP</a></td>
      <td>Live DOM, CSS &amp; Layout</td>
      <td>Hallucinating CSS cascade fixes without inspecting computed styles, line measure, or console errors.</td>
      <td><code>claude mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest</code></td>
    </tr>
    <tr>
      <td>03</td>
      <td><a href="https://chromologium.com/connect">Chromologium MCP</a></td>
      <td>Materials, Optics, OKLCH &amp; Behaviour</td>
      <td>Weightless SaaS defaults, hex guessing, dark-mode halation, muddy RGB mixes, 320px blowouts, and conflating cited references with UI roles.</td>
      <td><code>claude mcp add --transport http chromologium https://chromologium.com/mcp</code></td>
    </tr>
    <tr>
      <td>04</td>
      <td><a href="https://playwright.dev">Playwright MCP</a></td>
      <td>Multi-Viewport Interaction &amp; Flow</td>
      <td>Shipping responsive overflows, broken focus traps, and untested stateful user journeys across viewports.</td>
      <td><code>claude mcp add playwright -- npx @playwright/mcp@latest</code></td>
    </tr>
    <tr>
      <td>05</td>
      <td><a href="https://storybook.js.org">Storybook MCP</a></td>
      <td>Component Contracts &amp; Variants</td>
      <td>Inventing duplicate UI primitives or hallucinating prop signatures that violate existing design system stories.</td>
      <td><code>claude mcp add storybook -- npx -y @storybook/mcp@latest</code></td>
    </tr>
    <tr>
      <td>06</td>
      <td><a href="https://ui.shadcn.com">shadcn/ui Registry MCP</a></td>
      <td>Accessible Headless Primitives</td>
      <td>Pulling stale training-cutoff Radix wrappers instead of exact, dependency-resolved registry source files.</td>
      <td><code>claude mcp add shadcn -- npx shadcn@latest mcp</code></td>
    </tr>
    <tr>
      <td>07</td>
      <td><a href="https://www.deque.com/axe/">Axe Accessibility MCP</a></td>
      <td>WCAG, ARIA &amp; A11y Tree</td>
      <td>Treating accessibility as prose review rather than running deterministic rule audits against the live accessibility tree.</td>
      <td><code>claude mcp add axe -- npx -y @axe-core/mcp@latest</code></td>
    </tr>
    <tr>
      <td>08</td>
      <td><a href="https://21st.dev">21st.dev Magic MCP</a></td>
      <td>Motion &amp; Layout Exploration</td>
      <td>Defaulting to generic SaaS hero templates when rapidly exploring interaction choreography and layout studies.</td>
      <td><code>claude mcp add magic -- npx -y @21st-dev/magic@latest</code></td>
    </tr>
    <tr>
      <td>09</td>
      <td><a href="https://svgl.app">SVGL</a> / <a href="https://iconify.design">Iconify MCP</a></td>
      <td>Canonical Vector Marks &amp; Icons</td>
      <td>Hallucinating broken SVG path coordinates for brand marks and system icons from token memory.</td>
      <td><code>claude mcp add iconify -- npx -y iconify-mcp@latest</code></td>
    </tr>
    <tr>
      <td>10</td>
      <td><a href="https://www.blender.org">Blender</a> / <a href="https://threejs.org">Three.js Scene MCP</a></td>
      <td>Spatial, Lighting &amp; 3D Scenes</td>
      <td>Blindly editing 3D transforms, camera frustums, and PBR shaders without live scene-graph inspection.</td>
      <td><code>claude mcp add blender -- uvx blender-mcp</code></td>
    </tr>
  </tbody>
</table>

<h2>The Adjacent Bench: Four Substitutions by Authoring Stack</h2>
<table>
  <thead>
    <tr>
      <th>MCP Server</th>
      <th>Stack Swap Rule</th>
      <th>What It Grounds</th>
      <th>Claude Code CLI Install</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><a href="https://penpot.app">Penpot MCP</a></td>
      <td>Swap for #01 Figma</td>
      <td>Native SVG/CSS Grid &amp; Flex layout trees and open design-token files.</td>
      <td><code>claude mcp add penpot -- npx -y @penpot/mcp-server@latest</code></td>
    </tr>
    <tr>
      <td><a href="https://www.framer.com">Framer MCP</a></td>
      <td>Swap for #01 / #08</td>
      <td>CMS collections, responsive stacks, and motion code-component overrides.</td>
      <td><code>claude mcp add --transport sse framer https://mcp.unframer.co/sse</code></td>
    </tr>
    <tr>
      <td><a href="https://webflow.com">Webflow MCP</a></td>
      <td>Swap for #01 / #05</td>
      <td>Designer variables, class combo hierarchies, and structured CMS schemas.</td>
      <td><code>claude mcp add webflow -- npx -y webflow-mcp-server@latest</code></td>
    </tr>
    <tr>
      <td><a href="https://context7.com">Context7 MCP (Upstash)</a></td>
      <td>Pair with #02 / #06</td>
      <td>Version-pinned Tailwind v4, CSS Anchor Positioning, View Transitions, and React 19 documentation.</td>
      <td><code>claude mcp add context7 -- npx -y @upstash/context7-mcp@latest</code></td>
    </tr>
  </tbody>
</table>

<h2>Quickstart Configuration: The Daily Design Triad (<code>.mcp.json</code>)</h2>
<p>Mount three servers first — intent (<a href="https://figma.com">Figma</a>), perceptual math (<a href="https://chromologium.com/connect">Chromologium</a>), and glass (<a href="https://developer.chrome.com/docs/devtools">Chrome DevTools</a>) — before adding specialized instruments per task:</p>
<pre><code class="language-json">{
  "mcpServers": {
    "figma": {
      "url": "https://mcp.figma.com/mcp"
    },
    "chromologium": {
      "url": "https://chromologium.com/mcp"
    },
    "chrome-devtools": {
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp@latest"]
    }
  }
}</code></pre>
<p>For chromatic error measurements, five epistemic MCP families beyond design, and non-linear semi-lattice systems architecture, read the companion curatorial monographs: <a href="https://chromologium.com/monographs/the-instrument-in-the-context-window">The Instrument in the Context Window (MONO-11)</a> and <a href="https://chromologium.com/monographs/the-constellation-and-the-monolith">The Constellation and the Monolith (MONO-12)</a>.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/ten-design-mcps">https://evelynabellamy.com/writing/ten-design-mcps</a>.</em></p>]]></content:encoded>
    </item>
    <item>
      <title>The Legibility Trap</title>
      <description>Why making things 'easy to understand' sometimes destroys the very thing worth understanding.</description>
      <link>https://evelynabellamy.com/writing/the-legibility-trap</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/the-legibility-trap</guid>
      <pubDate>Thu, 15 Jan 2026 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <media:content url="https://evelynabellamy.com/images/journal/territory.webp" medium="image" type="image/webp" />
      <content:encoded><![CDATA[<figure><img src="https://evelynabellamy.com/images/journal/territory.webp" alt="Illustrated graphite mountains crossed by translucent blue map planes." /><figcaption>01 / Map and territory</figcaption></figure>

<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2026-01-15 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/the-legibility-trap">Read on evelynabellamy.com</a></p>
<blockquote><p>Why making things 'easy to understand' sometimes destroys the very thing worth understanding.</p></blockquote>

<p>There's a reflex in product work — in all organizational work, really — to make things legible. Readable. Parsable at a glance. We flatten complexity into dashboards, reduce nuance to bullet points, compress living systems into two-by-two matrices. And most of the time, this is fine. Necessary, even. You can't run a company on vibes alone.</p>
<p>But there's a point where the drive toward legibility stops serving clarity and starts destroying it. Where the map becomes so clean that it no longer resembles the territory. Where the summary is so crisp that it says nothing at all.</p>

<h2>The Seduction of the Simple</h2>
<p>James C. Scott wrote about this in Seeing Like a State — how governments, in their drive to make populations legible, would flatten local knowledge, rename villages, impose grid patterns on organic cities. The goal was always efficiency, control, understanding. The result was often catastrophe.</p>
<p>We do the same thing in product design, just at a smaller scale. We take a rich, complicated user behavior and reduce it to a funnel. We take a team's messy, productive creative process and flatten it into a linear workflow tool. We take a relationship between a person and a piece of software and reduce it to a satisfaction score.</p>
<p>Each of these translations loses something. And what's lost is usually the thing that matters most.</p>

<h2>What Gets Lost</h2>
<p>The problem isn't simplification itself. Simplification is a tool, and like any tool, it has appropriate and inappropriate uses. The problem is when we mistake the simplified version for the real thing. When the dashboard becomes reality and the thing the dashboard measures becomes an afterthought.</p>
<p>I've watched teams optimize metrics that were supposed to be proxies for user happiness, long after those metrics stopped correlating with anything meaningful. The legible version — the number going up — replaced the illegible reality of whether people actually found the product useful.</p>
<p>This is the trap. Legibility feels like understanding, but it's often just the appearance of understanding. And the appearance of understanding is more dangerous than honest confusion, because it stops you from asking questions.</p>

<h2>Holding Complexity Without Drowning in It</h2>
<p>So what's the alternative? Not illegibility for its own sake. Not complexity worship. Not the kind of intellectual posturing that refuses to explain anything clearly because clarity feels beneath you.</p>
<p>The alternative is what I'd call honest translation. It's the practice of simplifying while acknowledging what you've left out. Of building dashboards that include a section for what they can't measure. Of writing summaries that name their own blind spots.</p>
<p>The best systems I've worked on do this. They present clean interfaces while maintaining access to the mess underneath. They give you the summary and the source material. They let you zoom in.</p>
<p>Making things easy to understand is valuable. But understanding itself — real understanding — is rarely easy. The best we can do is build systems that respect that tension instead of pretending it doesn't exist.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/the-legibility-trap">https://evelynabellamy.com/writing/the-legibility-trap</a>.</em></p>]]></content:encoded>
    </item>
    <item>
      <title>Systems That See</title>
      <description>On the difference between platforms that surface patterns and those that impose them.</description>
      <link>https://evelynabellamy.com/writing/systems-that-see</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/systems-that-see</guid>
      <pubDate>Mon, 01 Dec 2025 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <media:content url="https://evelynabellamy.com/images/journal/seeing.webp" medium="image" type="image/webp" />
      <content:encoded><![CDATA[<figure><img src="https://evelynabellamy.com/images/journal/seeing.webp" alt="Overlapping blue lenses frame different views of an illustrated landscape." /><figcaption>02 / Ways of seeing</figcaption></figure>

<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2025-12-01 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/systems-that-see">Read on evelynabellamy.com</a></p>
<blockquote><p>On the difference between platforms that surface patterns and those that impose them.</p></blockquote>

<p>Every platform has an opinion about reality. Some are honest about it. Most aren't.</p>
<p>When you open a social media feed, you're not seeing the world. You're seeing the world as filtered through an algorithm that has its own goals — engagement, time-on-site, ad impressions. The platform isn't showing you what's happening. It's showing you what it thinks will keep you scrolling.</p>
<p>This isn't a conspiracy. It's just architecture. Every system that organizes information makes choices about what to foreground and what to hide, what counts as signal and what gets filed as noise. The question isn't whether a platform shapes your perception — it always does. The question is how honestly it does so.</p>

<h2>Surfacing vs. Imposing</h2>
<p>There's a meaningful difference between a system that surfaces patterns in your data and one that imposes patterns on it.</p>
<p>A system that surfaces patterns says: here's what we see in the information you've given us. Here are clusters, trends, anomalies. Make of them what you will. The interpretation is yours.</p>
<p>A system that imposes patterns says: here's what your data means. Here's what you should do about it. Here's the narrative, pre-assembled, ready to consume.</p>
<p>The first approach treats the user as an intelligent agent capable of judgment. The second treats them as a consumer of conclusions. Both have their place — sometimes you genuinely need someone (or something) to just tell you the answer. But the second approach is far more dangerous when misapplied, because it substitutes the system's judgment for yours without making that substitution visible.</p>

<h2>The Mirror and the Projector</h2>
<p>I think of it as the difference between a mirror and a projector. A mirror shows you yourself, with all the imperfections and contradictions intact. A projector shows you an image of what someone else wants you to see.</p>
<p>The best tools I've used are mirrors. A well-designed analytics dashboard doesn't tell you your product is good or bad — it shows you what's happening and lets you draw your own conclusions. A good research synthesis doesn't flatten conflicting findings into a single story — it holds the contradictions and helps you sit with them.</p>
<p>The worst tools are projectors that pretend to be mirrors. They present manufactured narratives as if they were objective observations. They give you the feeling of insight without the work of actually understanding.</p>

<h2>Building Honest Systems</h2>
<p>When I'm building something — a platform, a design system, an editorial tool — I try to keep this distinction front of mind. What am I surfacing? What am I imposing? Where am I making editorial choices that I'm presenting as neutral observations?</p>
<p>This doesn't mean building systems without opinions. Opinionated software is often the best software. It means being transparent about where those opinions live. It means building systems that show their work.</p>
<p>A platform that sees clearly is one that helps you see clearly too. Not by simplifying reality into digestible chunks, but by giving you better tools to navigate complexity on your own terms. That's a harder thing to build. It's also a more honest one.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/systems-that-see">https://evelynabellamy.com/writing/systems-that-see</a>.</em></p>]]></content:encoded>
    </item>
    <item>
      <title>The Warmth in the Machine</title>
      <description>How to design technology that feels human without pretending to be human.</description>
      <link>https://evelynabellamy.com/writing/the-warmth-in-the-machine</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/the-warmth-in-the-machine</guid>
      <pubDate>Mon, 20 Oct 2025 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <media:content url="https://evelynabellamy.com/images/journal/warmth.webp" medium="image" type="image/webp" />
      <content:encoded><![CDATA[<figure><img src="https://evelynabellamy.com/images/journal/warmth.webp" alt="A sage botanical sprig against delicate geometric construction lines." /><figcaption>03 / Organic and constructed</figcaption></figure>

<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2025-10-20 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/the-warmth-in-the-machine">Read on evelynabellamy.com</a></p>
<blockquote><p>How to design technology that feels human without pretending to be human.</p></blockquote>

<p>There's a particular brand of corporate warmth that makes my skin crawl. You know it: the chatbot that says &quot;I'm sorry you're feeling frustrated!&quot; The loading screen that says &quot;Hang tight, we're working on it!&quot; The error message that says &quot;Oops! Something went wrong.&quot;</p>
<p>It's warmth without substance. Friendliness as a design pattern. And it's everywhere — this idea that making technology feel more human means making it mimic human emotional expression. As if adding an exclamation point and an emoji to an error message makes the error less frustrating.</p>

<h2>The Uncanny Valley of Personality</h2>
<p>The problem with fake warmth isn't that it's warm. It's that it's fake. And people can tell. We've spent hundreds of thousands of years evolving exquisitely sensitive bullshit detectors, and they fire reliably when a piece of software tries to pretend it has feelings.</p>
<p>This creates an uncanny valley of personality. A system that's straightforwardly mechanical feels honest — you know where you stand. A system that genuinely understands your context and responds thoughtfully feels magical. But a system that performs understanding without having it feels creepy. It's the technological equivalent of a stranger smiling too wide.</p>
<p>The instinct behind it is right: technology should feel humane. Interfaces should treat people with respect. Interactions should be pleasant. But humane isn't the same as human. And treating people with respect doesn't require pretending to be a person.</p>

<h2>Warmth Through Competence</h2>
<p>The warmest technology I've ever used didn't try to be my friend. It was just exceptionally good at what it did.</p>
<p>A text editor that saves automatically so I never lose work. A mapping app that quietly reroutes around traffic without asking permission. A search engine that understands what I meant, not just what I typed. None of these feel &quot;friendly&quot; in the way a chatbot does. They feel caring. They feel like someone thought carefully about what I'd need and built it without fanfare.</p>
<p>This is warmth through competence. It's the design equivalent of a bartender who refills your water without you asking. No performance, no personality, just attentiveness translated into action.</p>

<h2>Designing for Dignity</h2>
<p>If there's a principle here, it might be this: design for dignity, not for personality.</p>
<p>Dignity means not wasting someone's time. Not making them repeat themselves. Not forcing them through a twelve-step process when a two-step process would do. Not showing them a cute animation when they're trying to accomplish something urgent.</p>
<p>Dignity means being reliable. Doing what you said you'd do. Failing gracefully when you can't. Being honest about limitations instead of hiding them behind cheerful copy.</p>
<p>Dignity means remembering that the person using your product has a life outside of it. That they came to accomplish something specific, and your job is to help them do it as painlessly as possible and then get out of the way.</p>
<p>That's real warmth. Not the simulated kind. Not the kind that says &quot;We're so glad you're here!&quot; The kind that demonstrates it by being genuinely, quietly useful.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/the-warmth-in-the-machine">https://evelynabellamy.com/writing/the-warmth-in-the-machine</a>.</em></p>]]></content:encoded>
    </item>
    <item>
      <title>Precision and Tenderness</title>
      <description>Notes on holding technical rigor and emotional intelligence in the same hand.</description>
      <link>https://evelynabellamy.com/writing/precision-and-tenderness</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/precision-and-tenderness</guid>
      <pubDate>Fri, 05 Sep 2025 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <content:encoded><![CDATA[<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2025-09-05 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/precision-and-tenderness">Read on evelynabellamy.com</a></p>
<blockquote><p>Notes on holding technical rigor and emotional intelligence in the same hand.</p></blockquote>

<p>There's a persistent myth in technical work that rigor and empathy are opposing forces. That being precise means being cold. That caring about people means being soft on standards. That you have to choose.</p>
<p>This is, to put it precisely, nonsense.</p>

<h2>The False Binary</h2>
<p>I've worked with engineers who wrote beautiful, meticulous code and treated their colleagues like obstacles to be optimized around. I've worked with people-focused leaders who were so worried about feelings that they couldn't deliver honest feedback. Both camps assumed they'd made the right trade-off. Both were wrong.</p>
<p>The best work I've been part of happened when both qualities were present — when the team held rigorous technical standards and also gave a damn about each other. When code reviews were thorough and kind. When deadlines were real but the humans meeting them were treated as humans, not resources.</p>
<p>This isn't a compromise. It's not splitting the difference between two extremes. It's recognizing that precision and tenderness are complementary qualities, not competing ones.</p>

<h2>Precision as Care</h2>
<p>If you think about it, precision is a form of care. When you name a variable clearly, you're caring about the person who'll read your code next. When you write thorough documentation, you're caring about future-you and everyone who comes after. When you test edge cases, you're caring about the user who'll encounter them.</p>
<p>Sloppy work isn't compassionate. It just defers the pain. Every shortcut, every &quot;we'll fix it later,&quot; every untested assumption — these aren't kindnesses. They're debts that someone else will pay.</p>
<p>The most caring thing you can do in technical work is be precise. Not rigid — there's a difference. Rigid means inflexible, rule-following for its own sake. Precise means thoughtful, intentional, willing to spend the extra hour getting something right because you care about the outcome and the people it affects.</p>

<h2>Tenderness as Strength</h2>
<p>And on the other side: tenderness isn't weakness. Being attentive to how people feel doesn't make you less rigorous. Asking someone how they're doing before diving into a technical review doesn't waste time — it creates the psychological safety that makes honest feedback possible.</p>
<p>I've seen teams where people were afraid to admit they didn't understand something, because the culture punished vulnerability. Those teams shipped bugs. They missed edge cases. Not because they lacked technical skill, but because the absence of tenderness made the precision impossible.</p>
<p>When people feel safe, they ask better questions. They flag concerns earlier. They admit mistakes before those mistakes compound. Tenderness, in this context, isn't a soft skill — it's infrastructure. It's the foundation that makes everything else work.</p>

<h2>Both Hands</h2>
<p>So I try to hold both. Technical rigor in one hand, emotional intelligence in the other. Precision in the architecture, tenderness in the process. High standards for the work, genuine care for the people doing it.</p>
<p>It's not always easy. There are moments when they feel like they're pulling in different directions — when being honest about a technical problem might hurt someone's feelings, when being kind might mean letting a substandard solution slide. But those moments are navigable. They require judgment, not trade-offs.</p>
<p>The work is better when both qualities are present. The people are better off too. And those two things, it turns out, are not in conflict.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/precision-and-tenderness">https://evelynabellamy.com/writing/precision-and-tenderness</a>.</em></p>]]></content:encoded>
    </item>
    <item>
      <title>What Strategy Actually Means</title>
      <description>Reclaiming a word that's been diluted into meaninglessness by overuse.</description>
      <link>https://evelynabellamy.com/writing/what-strategy-actually-means</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/what-strategy-actually-means</guid>
      <pubDate>Fri, 18 Jul 2025 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <content:encoded><![CDATA[<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2025-07-18 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/what-strategy-actually-means">Read on evelynabellamy.com</a></p>
<blockquote><p>Reclaiming a word that's been diluted into meaninglessness by overuse.</p></blockquote>

<p>Everyone has a strategy now. Your social media presence has a strategy. Your lunch break has a strategy. Your strategy has a strategy. The word has been used so freely and so loosely that it's come to mean something like &quot;a plan&quot; or &quot;a thing we thought about for more than five minutes&quot; or, most commonly, &quot;a document we made to justify a decision we'd already made.&quot;</p>
<p>This is a problem. Not a semantic one — language evolves, words shift meaning, that's fine. It's a practical problem. When strategy means everything, it means nothing. And when it means nothing, we lose the ability to talk about the very real and important thing that strategy, properly understood, actually refers to.</p>

<h2>What It's Not</h2>
<p>Strategy is not a plan. A plan is a sequence of steps. A strategy is the logic that determines which steps to take and which to skip.</p>
<p>Strategy is not a goal. &quot;We want to grow revenue by 20%&quot; is a goal. Strategy is how you're going to do it — and, more importantly, what you're going to not do in order to make it possible.</p>
<p>Strategy is not a mission statement. &quot;We empower individuals to reach their full potential&quot; is a sentiment, not a strategy. You can't disagree with it, which is how you know it's not strategic. Real strategy involves choices that reasonable people could disagree with.</p>
<p>Strategy is not a slide deck. It might be communicated via slide deck, but the deck is the vehicle, not the cargo.</p>

<h2>What It Is</h2>
<p>Strategy is a theory of advantage. It's your best current explanation for how you're going to succeed, given the constraints you're operating under. It's a bet — an informed, deliberate bet about where to focus your limited resources to achieve disproportionate impact.</p>
<p>The key word is focus. Strategy is less about what you're going to do and more about what you're going to stop doing. Every meaningful strategy involves sacrifice — real sacrifice, not the rhetorical kind where you &quot;deprioritize&quot; something you weren't doing anyway.</p>
<p>Good strategy is painful. It requires saying no to good ideas in service of great ones. It requires killing projects that people are emotionally attached to. It requires accepting that you can't do everything and being explicit about what you're leaving on the table.</p>

<h2>The Test</h2>
<p>Here's a simple test for whether something is a strategy: can you imagine a reasonable person disagreeing with it? If not, it's probably not strategic. &quot;We're going to focus on quality&quot; isn't a strategy because no one would argue for the alternative. &quot;We're going to ship faster at the expense of polish&quot; is a strategy because it involves a genuine trade-off that someone could reasonably object to.</p>
<p>Another test: does it help you make decisions? A good strategy makes hundreds of downstream decisions easier, because it provides a framework for saying yes or no. If your &quot;strategy&quot; doesn't help you decide what to do on a Tuesday afternoon when two priorities conflict, it's not a strategy. It's a poster.</p>

<h2>Why It Matters</h2>
<p>I care about this because I've watched teams drown in activity — shipping features, running experiments, launching initiatives — without any strategic coherence. Everyone busy, no one effective. Lots of motion, no movement.</p>
<p>Strategy, properly practiced, is the cure for this. Not because it makes everything easy, but because it makes the hard decisions visible and deliberate instead of invisible and accidental. It forces you to say, out loud, what you believe and why. And that, at minimum, gives everyone else the information they need to tell you you're wrong.</p>
<p>Which, if you're doing strategy right, should happen fairly often.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/what-strategy-actually-means">https://evelynabellamy.com/writing/what-strategy-actually-means</a>.</em></p>]]></content:encoded>
    </item>
    <item>
      <title>The Adjacent Possible</title>
      <description>Why the best innovations are one step away from what already exists.</description>
      <link>https://evelynabellamy.com/writing/the-adjacent-possible</link>
      <guid isPermaLink="true">https://evelynabellamy.com/writing/the-adjacent-possible</guid>
      <pubDate>Mon, 02 Jun 2025 12:00:00 GMT</pubDate>
      <dc:creator>Evelyna Bellamy</dc:creator>
      <category>Translation, Actually</category>
      <content:encoded><![CDATA[<p><strong>Translation, Actually</strong> · <strong>Published:</strong> 2025-06-02 · <strong>Author:</strong> <a href="https://evelynabellamy.com">Evelyna Bellamy</a> · <a href="https://evelynabellamy.com/writing/the-adjacent-possible">Read on evelynabellamy.com</a></p>
<blockquote><p>Why the best innovations are one step away from what already exists.</p></blockquote>

<p>Stuart Kauffman coined the term &quot;the adjacent possible&quot; to describe the set of all things that are one step away from what currently exists. In biology, it's the set of molecules that could form from the molecules already present. In technology, it's the set of inventions that could be built from the components already available. In creative work, it's the set of ideas that are reachable from where you currently stand.</p>
<p>It's a concept I think about constantly, because it explains something that's otherwise hard to articulate: why the best innovations feel simultaneously surprising and inevitable.</p>

<h2>The Myth of the Leap</h2>
<p>We love stories about revolutionary breakthroughs. The lone genius who sees what no one else can see. The startup that disrupts an entire industry with a radical new idea. The moment of inspiration that changes everything.</p>
<p>These stories are almost always wrong, or at least heavily edited. When you look at the history of any significant innovation, what you find is not a leap but a series of steps. Each step was possible because of what came before. Each step opened up new possibilities that weren't accessible before.</p>
<p>The iPhone didn't come from nowhere. It came from decades of touchscreen research, mobile computing development, miniaturized sensors, and cellular infrastructure. It was, in 2007, adjacent to what already existed. That doesn't make it less impressive — it makes it more impressive, because it means someone looked at the landscape of what was possible and saw a combination that no one else had assembled.</p>

<h2>Why This Matters for Building Things</h2>
<p>If you're trying to build something new — a product, a platform, a company, a creative work — the adjacent possible is the most useful constraint you can apply to your thinking.</p>
<p>It means asking: what's one step away from what I already have? What could I build with the components that are already available to me? What combination of existing things would create something genuinely new?</p>
<p>This isn't settling for incrementalism. The adjacent possible can be enormous. It's not about making small improvements to existing things — it's about seeing connections between existing things that create something qualitatively different. The web browser was adjacent to hypertext and networking. Wikipedia was adjacent to the web browser and the tradition of encyclopedias. Neither was incremental. Both were adjacent.</p>

<h2>The Discipline of One Step</h2>
<p>What I love about this framework is that it's simultaneously ambitious and disciplined. It encourages you to think big — the adjacent possible is vast, full of unexplored combinations — while keeping you grounded in reality. You're not fantasizing about what might be possible in ten years with unlimited resources. You're looking at what's possible right now, with what you have.</p>
<p>In practice, this means paying deep attention to what already exists. To the tools, technologies, behaviors, and cultural shifts that are already in motion. The best innovations don't fight the current — they ride it in a direction no one expected.</p>
<p>It also means being suspicious of ideas that require too many things to change simultaneously. If your product needs users to adopt new hardware, learn new mental models, and change established habits all at the same time, you're probably not in the adjacent possible. You're in the distant possible, which is a fancy way of saying you're probably too early.</p>

<h2>One Step from Here</h2>
<p>I try to apply this thinking to my own work. When starting a new project, I ask: what do I already know how to build? What tools do I already have? What skills, relationships, and insights am I already sitting on? And then: what could I combine from that inventory to create something new?</p>
<p>The answers are usually more interesting than whatever I'd dream up from scratch. Because the adjacent possible isn't limited by my imagination — it's expanded by it. The more I know about what exists, the more combinations I can see. The more combinations I can see, the larger my adjacent possible becomes.</p>
<p>Innovation, it turns out, isn't about seeing further than everyone else. It's about seeing more clearly what's right in front of you.</p>

<hr />
<p><em>Published by <a href="https://evelynabellamy.com">Evelyna Bellamy</a> in <a href="https://evelynabellamy.com/writing">Writing</a>. Canonical URL: <a href="https://evelynabellamy.com/writing/the-adjacent-possible">https://evelynabellamy.com/writing/the-adjacent-possible</a>.</em></p>]]></content:encoded>
    </item>
  </channel>
</rss>
