<?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 Mom Who Vibe Codes on Medium]]></title>
        <description><![CDATA[Stories by Mom Who Vibe Codes on Medium]]></description>
        <link>https://medium.com/@momwhovibecodes?source=rss-3a46b6e75ae6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*4di3YiqwTx_xTCWy5paEIw.png</url>
            <title>Stories by Mom Who Vibe Codes on Medium</title>
            <link>https://medium.com/@momwhovibecodes?source=rss-3a46b6e75ae6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 08:11:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@momwhovibecodes/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[Mom Who Vibe Codes — Mother’s Day Edition]]></title>
            <link>https://medium.com/@momwhovibecodes/mom-who-vibe-codes-mothers-day-edition-f43d0dce8848?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/f43d0dce8848</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 10 May 2026 22:52:00 GMT</pubDate>
            <atom:updated>2026-05-10T22:52:00.482Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s Mother’s Day. I am living up all three dimensions of mom who vibe codes and doing all the things that bring me joy.</p><p>⚽ First up, I took my 13 year old to his 9am game and watched him step onto the pitch with his calm energy. I love watching him move the ball with intention and speed, the swiftness in his decision-making, and how he can change the game dynamic with a few simple moves. His team wrapped the game with a 3–1 win.</p><p>🧇 After a hard fought victory, I made the boys’ favorite Sunday brunch: Belgian waffles with homemade chai latte. What is better than sharing delicious food together with my favorite people?</p><p>💻 I then decided that it is time for me to learn how to vibe design — I am not a UX designer and I don’t have the vocabulary to describe a design to coding agents. But agents are great teachers. So I asked Claude what design skills it has and how I can improve my design language. It instantly helped me understand how to break down a design into different aspects and how to talk about each:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rz-Rg4Y200ijrXpBjzIsQQ.png" /></figure><p>With this new found knowledge, I proceeded to give my AI soccer coach app a facelift. It became clear to me that the primary user is parents of the players. I want the app design to reflect that with a premium, polished brand appearance — or as Claude called it, a “cinematic effect.” Here are just a few of our iterations:</p><h3>📱 Original</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/942/1*fr-dTMDkmvNPzCoJQirszQ.png" /></figure><h3>📱 V1</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xDpSMLMiyygr0uv_M3OgGA.png" /></figure><h3>📱 V2</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PE6CEWUfd6-L5eyBzBf0mA.png" /></figure><h3>📱 V7</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AE38BmDxsRoSJwfzTRHuUg.png" /></figure><p>While Claude was designing, I was vibing — as in baking up some Japanese milk bread 🍞. Vibing and baking went on for about 90 minutes, and I came out with a winner design (v13!) and two golden-brown loaves.</p><p>📚 A few new words I learned today in vibe designing:</p><p>- <strong>Visual hierarchy </strong>— what your eye sees first, second, third. If everything is loud, nothing is.</p><p>- <strong>Affordance</strong>— does an element <em>*look*</em> like what it does? A button should look pressable; a card should look tappable.</p><p>- <strong>Density</strong> — how packed the screen is. Premium dark UIs are usually low-density with generous whitespace.</p><p>- <strong>Brand expression</strong> — how distinctive and intentional a design feels vs. looking like a template. The line between “generic AI-generated” and “real product.”</p><p>- <strong>Cinematic</strong> — atmospheric, lit-from-within, premium-dark. Light does the work, not color.</p><p>- <strong>Micro-interaction</strong> — a small motion moment that carries meaning. Our 🔥 emoji blooms from the bottom when Hype mode activates; 😌 slowly breathes when Calm activates.</p><p>- <strong>Variants must be distinct</strong> — when picking between N options, each has to be in a meaningfully different category. Subtle parameter tweaks aren’t real choices.</p><p>- <strong>Saturation</strong>— color intensity. A color that pops as a small accent (border, dot, glow) can feel overwhelming when used as a full fill.</p><p>- <strong>Use visual as the cue</strong> — when a chip glows and pulses, you don’t need a “Recommended” label. The design IS the message.</p><p>To top it all off, my kids gave me the best card:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*i_if-GjmECLbzaFPuxmohQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3rKMgZIKq1_IVNUKvbz1rQ.jpeg" /></figure><p>✅ <strong>Mom</strong> — checked</p><p>🎨 <strong>Vibe</strong> — watching my kids play, making food, enjoying our time together</p><p>💻 <strong>Code</strong> — vibe designed my way through a facelift</p><p>Happy Mother’s Day to all the moms out there!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f43d0dce8848" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Turning Tournament Sideline Chats Into A New Feature: Building a College Recruitment Guide in One…]]></title>
            <link>https://medium.com/@momwhovibecodes/turning-tournament-sideline-chats-into-a-new-feature-building-a-college-recruitment-guide-in-one-e97d727cd5f1?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/e97d727cd5f1</guid>
            <category><![CDATA[soccer]]></category>
            <category><![CDATA[parenting]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 22 Mar 2026 21:14:04 GMT</pubDate>
            <atom:updated>2026-03-22T21:14:04.834Z</atom:updated>
            <content:encoded><![CDATA[<h3>Turning Tournament Sideline Chats Into A New Feature: Building a College Recruitment Guide in One Weekend ⚽</h3><p><em>How I turned my real-life discovery into a feature that helps families navigate the black box of college soccer recruitment</em></p><h3>The Problem I Casually Stumbled Upon 🤔</h3><p>I was in Virginia for the Jefferson Cup with my 13-year old. I never knew what to expect from these tournaments but to my surprise, I had lots of bonding time with my fellow soccer moms on the sideline. One topic piqued my curiosity — parents with high school aged kids are starting to navigate the college recruitment process. It dawned on me that college soccer recruitment is a black box. As a parent with a rising middle-schooler playing in MLS Next, I am clueless yet feel the urge to get educated. With some basic google searches, the information exists, but it’s scattered across forums, buried in NCAA PDFs, and gatekept by expensive consultants. First-time families like mine are left piecing together a puzzle with no picture on the box. 🧩</p><p>So I did what any product manager would do: I built a solution. 🛠️</p><h3>What I Shipped Today 🚀</h3><p>I added two new features to my AI Soccer Coach app:</p><h3>1. The Recruitment Readiness Quiz ✅</h3><p>Seven quick questions that tell families where they stand:</p><ul><li>What’s your child’s age and club level?</li><li>Do you have a highlight video?</li><li>Have you registered with the NCAA Eligibility Center?</li></ul><p>The quiz spits out a personalized “readiness score” and concrete next steps. No fluff. No sales pitch for a $$$ recruiting service. 🙅‍♀️</p><h3>2. Coach Fabian’s Recruitment Mode 💬</h3><p>My AI coach (who already helps kids with pre-game nerves and video analysis) now has a recruitment brain. Parents and kids can ask questions like:</p><ul><li><em>“What’s the difference between D1 and D3 scholarships?”</em></li><li><em>“My son is a sophomore. Is it too late to start?”</em></li><li><em>“What should be in a highlight video?”</em></li></ul><p>And get clear, jargon-free answers. With bullet points. Because walls of text help no one.</p><h3>Key Decisions Before Any Code is Written 💡</h3><p>I asked Claude to keep in mind that we are building for both parents and kids 👨‍👩‍👧. Our problem to solve is to help our personas navigate the non-linear, stressful recruitment process by breaking it down with clear, personalized actions. I also instructed Claude to differentiate the solution from traditional consulting services. No fancy, 15-step prompts, just simple two line questions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pehKB_7F68BeYrijnlWdtw.png" /></figure><p>Another thing I asked Claude to think about is distribution strategy 📲 — how do I get the word out and get real users to try this feature? Without a distribution strategy, the feature is just going to collect dust. Claude being the strong thinking partner helped me land on building a shareable quiz with a hook into a chat session with Coach Fabian.</p><h3>The Building Part: Iterating on UX in Real-Time 🔄</h3><p><strong>Problem 1: Responses were getting cut off.</strong> ✂️</p><p>Coach Fabian would start explaining scholarships and then… just stop mid-sentence. The culprit? I had max_tokens set to 150 (fine for quick pep talks, not for explaining D1 vs D3 scholarships). Quick fix: bumped it to 500 for recruitment mode. ✅</p><p><strong>Problem 2: Walls of text.</strong> 📜</p><p>Even with more tokens, the responses were dense paragraphs. Nobody wants to read that on their phone at a tournament. I added format rules to the prompt:</p><ul><li>Use bullet points, not paragraphs</li><li>Emojis only in section headers (🏆 D1 | 🎯 D2 | 📚 D3)</li><li>Keep responses under 150 words before follow-up topics</li></ul><p><strong>Problem 3: Conversations hit dead ends.</strong> 🛑</p><p>Users would ask a question, get an answer, and then… stare at a blank input field. What now? I added follow-up topic chips that appear after every response:</p><blockquote><strong><em>Want to know more?</em></strong><em> • How scholarships work • When to contact coaches • What goes in a highlight video</em></blockquote><p>Claude extracted these from the response content, with fallback topics if extraction failed. Small detail, big UX improvement. ✨</p><p><strong>Problem 4: Random money emoji appearing.</strong> 💰</p><p>Coach Fabian kept dropping 💰 emojis in weird places. The fix? Simplified the emoji guidance to ONLY use them in section headers. Constraints &gt; creativity when you’re shipping fast. 🎯</p><h3>Building an Eval (Because Vibes Need to Be Grounded) 🧪</h3><p>I needed to know if Coach Fabian was actually giving good advice. But I didn’t want to manually test every response. So I asked Claude to help me build an evaluation system.</p><p><strong>Step 1: Claude generated 10 golden questions</strong> 📝 covering basics, edge cases, and common mistakes:</p><ul><li>“How do scholarships work?”</li><li>“My daughter is 16 and just started thinking about college soccer. Is it too late?”</li><li>“What’s JUCO?”</li></ul><p><strong>Step 2: Claude wrote the eval script</strong> 🖥️ that hits my API, checks responses for required keywords, and flags violations.</p><p><strong>Step 3: Run, fail, iterate.</strong> 🔁</p><p>First run: 30% pass rate. 😬</p><p>The eval caught real problems — Coach Fabian wasn’t mentioning JUCO (a great path to full scholarships) or explaining that players need to show “both feet” in highlight videos.</p><p>I fed the failures back to Claude, we updated the prompt, ran again: 60%. 🤔</p><p>Then we made the eval smarter. The word “too late” was flagging even when the response said “it’s NEVER too late.” Claude added context-aware violation detection:</p><ul><li>✅ “It’s never too late” → PASS</li><li>❌ “It might be too late” → FAIL</li></ul><p>Final run: 80% pass rate. 📈</p><p>The whole eval loop — from “I want to test this” to “80% passing” — took maybe 20 minutes. That’s the power of building with a coding agent that can write its own tests. 🤖</p><p>Total build time: ~1.5 hours across a weekend afternoon. ⏱️</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c-CWOA9JjMBehNXW4XbX4A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/514/1*VNkb-vmJu3bZyIGrzK4p6Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/578/1*grMaIofeQmbsHMooFAbyFw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/522/1*ZTKMTswxwgw9PYwUAg_PpQ.png" /></figure><h3>What’s Next 👀</h3><p>This is a side project, so I’m shipping small and learning:</p><ol><li><strong>User testing</strong> 🧑‍🤝‍🧑 — Getting real soccer parents to try the quiz and chat</li><li><strong>More eval coverage</strong> 📊 — Adding questions about showcases, coach outreach, and highlight videos</li><li><strong>Maybe video analysis for recruiting?</strong> 📹 — Imagine uploading a clip and getting feedback on what college coaches look for</li></ol><h3>The Real Takeaway 💡</h3><p>The best product ideas come from your constant discovery through real life conversations. Key decisions like core problems to solve, differentiation, and distribution strategy are important to shape the building process.</p><p>I didn’t set out to build a recruitment tool. I was just a curious soccer mom who wanted answers. Now I have an AI that gives them to me — and maybe helps a few other families along the way. 🙏</p><p>If you’re navigating college soccer recruitment and want to try it: <a href="https://ai-soccer-coach.vercel.app/quiz">ai-soccer-coach.vercel.app/quiz</a> 👆</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e97d727cd5f1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From “Keep Your Head Up” to Building 3D Worlds]]></title>
            <link>https://medium.com/@momwhovibecodes/from-keep-your-head-up-to-building-3d-worlds-b6aba678ca91?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6aba678ca91</guid>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 15 Mar 2026 22:33:59 GMT</pubDate>
            <atom:updated>2026-03-15T23:24:25.638Z</atom:updated>
            <content:encoded><![CDATA[<p>Last week I built a video analysis feature for my AI soccer coach app. It dawned on me that an LLM cannot see images or videos the way I see — the innate spatial context I have to make sense of what happened and who did what. I didn’t have the words for it at the time, but as I read Dr. Fei-Fei Li’s <a href="https://drfeifei.substack.com/p/from-words-to-worlds-spatial-intelligence">Substack</a>, I learned this concept is called “Spatial Intelligence.”</p><p>This week I fell down a rabbit hole about spatial intelligence and ended up creating a 3D preview of my summer vacation.</p><p>Let me explain. 🐇</p><h3>🤦🏻‍♀️ Image tokens are not the same as human eyes</h3><p>The tech behind my video analysis feature sends Claude still images extracted from video — I tuned it to send a fixed number of frames. The model processes these tokens and generates analysis based on my prompt. This is far from how we see and process the world: for the same video, Claude received 30 frames while my eyes processed 900–1,800 frames of continuous visual information, plus all the relational context.</p><p>Dr. Fei-Fei Li’s argument: LLMs are brilliant at abstract text work but fundamentally disconnected from physical reality. They can’t estimate distance, rotate objects mentally, or predict physics. They manipulate symbols without truly comprehending space.</p><p>This is exactly what I experienced:</p><h4>Claude saw pixels. I saw a kid playing soccer.</h4><h3>🌍 So I Tried Marble</h3><p>Dr. Fei-Fei Li isn’t just writing about spatial intelligence — she’s building it. Her company World Labs released <a href="https://marble.worldlabs.ai/">Marble</a>, a tool that generates 3D worlds from text and image prompts.</p><p>I had to try it. What better way to preview one of our summer vacation spots in Helsinki than a 3D walkthrough?</p><p>Now, I had no experience or skill making 3D worlds. So I enlisted Claude Code to create a step-by-step guide. Here’s the prompt we crafted:</p><p>```</p><p>Suomenlinna sea fortress, Helsinki, late July morning, 9am.</p><p>Soft golden sunlight casting long shadows across weathered</p><p>stone bastion walls. 18th century bronze cannons lined along</p><p>the ramparts. Dewy grass between cobblestone paths. Pink</p><p>granite rocks meeting the water’s edge. Peaceful, uncrowded,</p><p>the fortress just waking up.</p><p>```</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bYtloxX9-WroYLXVmCSsgQ.png" /><figcaption>Marble interface</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EHGSzi2pHRtOHe_zYwtXgA.png" /><figcaption>3D world generating</figcaption></figure><p>Six minutes later: a 3D world I could fly through.</p><h3>🔬 What I Learned (The Honest Version)</h3><p>It’s impressive… from the right angle</p><p>The aerial view looked like a soft watercolor painting. Beautiful, but blurry. When I navigated down to “ground level” — standing on the fortress walls looking out at the Baltic Sea — the detail improved dramatically.</p><p>This makes sense. Through Claude, I learned that Marble uses Gaussian splats (millions of tiny 3D ellipsoids) to render the scene. From far away, you’re looking at a cloud of colored dots. Up close, they resolve into textures.</p><h4>The editing tools aren’t what I expected 🙃</h4><p>I spent 10 minutes trying to “touch up” a blurry section before realizing Studio mode is for <em>composing multiple worlds together</em>, not editing detail within one.</p><p>Lesson: RTFM. Or at least skim it.</p><h4>The mental model shift is real</h4><p>This isn’t image generation. You’re not creating a picture — you’re creating a <em>place</em>. I could walk through my Suomenlinna, look around corners, see how the morning light would fall on the fortress walls.</p><p>That’s… different. And kind of magical for trip planning. ✨</p><h3>💡 The Bigger Picture</h3><p>What struck me most: the gap between “seeing images” and “understanding space” is exactly what Dr. Fei-Fei Li described.</p><p>My video analysis AI couldn’t perceive motion through time. But Marble is trying to solve the inverse problem — generating coherent 3D space that you can move through.</p><p>One tool is blind to spatial reality. The other is trying to create it.</p><p>Both are limited in different ways. But experiencing them side by side — the Claude that can’t see a kid’s head movement, and the Marble that can generate a walkable fortress — felt like seeing two different frontiers of AI.</p><h3>🗺️ Is This the New Frontier?</h3><p>Making sense of our place in the world is one of our unique abilities as humans. When AI acquires this skill, how will it change the way we interact with it and the world?</p><p>Will my kids grow up with AI coaches that can watch them play sports and actually <em>understand</em> what happened? Will architects walk clients through buildings that don’t exist yet?</p><p>And what about healthcare? Imagine an oncologist showing a patient not just scans and charts, but a 3D visualization of their own body — here’s the tumor, here’s how the radiation will target it, here’s what we expect to see in three months. The abstract becomes tangible. The terrifying becomes something you can look at, walk around, understand.</p><p>Right now, patients hear words like “tumor reduction” and “treatment response” and try to imagine what that means inside their own body. What if they could <em>see</em> it? What if spatial AI could turn a frightening unknown into something visible, navigable, even hopeful?</p><p>That’s not just a new technology. That’s a new way of understanding our own life.</p><p>While we ponder what this new frontier means for our future, if you are interested, try playing with my marble creation or create one yourself: <a href="https://marble.worldlabs.ai/world/e21c7cca-5baf-419a-ab75-7c29d8fc2d0a*"><em>https://marble.worldlabs.ai/world/e21c7cca-5baf-419a-ab75-7c29d8fc2d0a</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b6aba678ca91" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Easy Cinnamon Raisin Bread]]></title>
            <link>https://medium.com/@momwhovibecodes/easy-cinnamon-raisin-bread-701730235a00?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/701730235a00</guid>
            <category><![CDATA[baking]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cooking]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 01 Mar 2026 22:34:26 GMT</pubDate>
            <atom:updated>2026-03-01T22:36:17.846Z</atom:updated>
            <content:encoded><![CDATA[<p>Last weekend I pair programmed with Claude Code to build a video analysis feature. While Claude was extracting video frames and debugging server errors, I made cinnamon raisin bread from scratch.</p><p>First time ever. Turned out delicious.</p><p>Here’s the thing: bread-making has a lot of waiting. Rise for 3 hours. Bake for 30 minutes. Cool for 20. You know what else has a lot of waiting? Watching an AI iterate through approaches until one works.</p><p>Perfect pairing. 🍞🤖</p><h3><strong>The Recipe</strong></h3><p>Adapted from <a href="https://asimplepalate.com/blog/cinnamon-raisin-artisan-bread/">A Simple Palate</a>. You’ll need a Dutch oven.</p><h4>Ingredients</h4><p>- 3 1/4 cups all-purpose flour</p><p>- 1 1/2 cups warm water</p><p>- 2 tsp active dry yeast</p><p>- 3 tbsp brown sugar</p><p>- 3/4 tsp salt</p><p>- 2/3 cup raisins</p><p>- 2 tsp cinnamon</p><h4>Instructions</h4><p>1. Mix the dough</p><p>Combine warm water, brown sugar, and yeast. Let it sit 5 minutes — the sugar feeds the yeast. Add flour, salt, cinnamon, and raisins. Mix until a shaggy dough forms. If it’s too dry, add water 1 tbsp at a time. Too sticky? Add flour.</p><p>2. Let it rise</p><p>Transfer to a greased bowl, cover with a damp towel, and let rise for 3+ hours until doubled. Pro tip: put it in your oven with just the light on (oven OFF). The light generates enough warmth.</p><p>3. Shape it</p><p>Once doubled, dump onto a floured surface and shape into a ball. Place on parchment paper.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e8mB1kSzFwJ-qWEDXKtqaQ.jpeg" /></figure><p>4. Preheat</p><p>Heat oven to 450°F with your Dutch oven inside.</p><p>5. Bake</p><p>Carefully transfer the dough (using the parchment) into the hot Dutch oven. Cover and bake 30 minutes. Remove lid and bake another 5–10 minutes until golden.</p><p>6. Wait (the hardest part)</p><p>Let it cool 20 minutes before slicing. I know. It smells amazing. But the inside is still cooking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uYBDsPDfJYCnUMBl3FbEtA.jpeg" /></figure><h3>What I Learned</h3><p>Making bread is way less intimidating than I thought. Most of the “work” is just… waiting. Which made it perfect for a coding session where I wanted to stay engaged but not stare at a screen the whole time.</p><p>While the dough rose, Claude figured out why FFmpeg was crashing. While the bread baked, Claude added safety guardrails to the AI coach. While it cooled, I reviewed the code and tested the feature.</p><p>Multitasking at its finest.</p><h3>The Verdict</h3><p>The bread? Crusty outside, soft inside, perfect amount of cinnamon. My kids devoured it.</p><p>The code? Also worked. Video analysis feature shipped to production.</p><p>10/10 would recommend AI pair programming + baking as a workflow. 🍞</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=701730235a00" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Coach Fabian Can Now See]]></title>
            <link>https://medium.com/@momwhovibecodes/coach-fabian-can-now-see-ec7f298086f6?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec7f298086f6</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[soccer]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 01 Mar 2026 21:45:35 GMT</pubDate>
            <atom:updated>2026-03-01T21:45:35.595Z</atom:updated>
            <content:encoded><![CDATA[<p>For the past two weekends, I added a feature that lets kids upload their game footage and get personalized coaching feedback. Sounds simple, right?</p><p>Plot twist: It wasn’t. And that’s exactly why I learned so much. 🎢</p><p>The pair programming with an AI was eye opening — I felt like having a senior engineer sitting next to me, patiently explaining <em>why</em> we’re doing things a certain way. And while Claude worked, I made cinnamon raisin bread for the first time (recipe in another post!)</p><h3>🤝 Lesson 1: Claude Code as an Architect, Not Just a Coder</h3><p>I told Claude I wanted to build a video analysis feature. Without further prompting, Claude presented a thorough plan before writing a single line of code.</p><p>It presented these three options:</p><h4>Option 1: Partner with Veo (Long-term)</h4><p>- Reach out to Veo about API access</p><p>- Position it as “we help your users get more value from recordings”</p><h4>Option 2: URL Paste (Test First)</h4><p>- Parent pastes a Veo URL into the app</p><p>- App fetches the video or extracts frames</p><h4>Option 3: Manual Upload (Simplest MVP)</h4><p>- Parent downloads a short clip from Veo</p><p>- Uploads it directly to the app</p><p>- Works today, no dependencies</p><p>Claude’s recommendation? <em>Start with Option 3</em> to validate that parents actually want video analysis. “Trying to integrate with Veo before you know people want this feature = premature optimization.”</p><p>🤯 I did not realize I was getting PM decision-making lessons too.</p><p>Then came the clarifying questions — presented in a structured flow:</p><p>- “What type of video clips will kids upload?” (Short highlights vs longer clips)</p><p>- “What focus areas matter most?”</p><p>- “What format should results take?”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*x-cjeQ5b49nEOahuyGveUQ.png" /><figcaption>scope planning</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/997/1*LksStUmB21ZbuuJDacp-Jw.png" /><figcaption>user flow drafting</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/995/1*wSRoA_2q-bEppNChty_5Mg.png" /><figcaption>UI planning</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1001/1*ETdVkDl5-57pA50rH4OhRg.png" /><figcaption>Implementation planning</figcaption></figure><p>This wasn’t an AI waiting for instructions. It was AI <strong>leading</strong> the technical planning while keeping me in the decision-making seat.</p><p><strong>The insight</strong>: Claude wasn’t just executing my vision. It was architecting the solution, presenting tradeoffs, and recommending the pragmatic path. Like having a tech lead who drafts the design doc and walks you through it before implementation.</p><h3>🔥 Lesson 2: The Build Was NOT a Straight Line</h3><p>If you’ve ever watched a coding tutorial, you know they make everything look easy. Record, edit out the mistakes, publish. Magic! ✨</p><p>Real life? Messier. Much messier.</p><h4>Attempt #1: FFmpeg in the Browser</h4><p>The plan: Use FFmpeg WebAssembly to extract frames from the video client-side.</p><p>The reality: “Failed to process video”</p><p>The FFmpeg WASM approach required loading massive files, had cross-origin issues, and crashed on my 14MB test video. I stared at a stuck loading spinner while the error message got cut off at the bottom of the screen. 😅</p><h4>Attempt #2: Native Browser APIs</h4><p>Claude pivoted: “Let me switch to a simpler, more reliable approach using the native HTML5 video element and canvas.”</p><p>Wait, we can just… do that? Use the browser’s built-in video player, seek to different timestamps, draw frames to a canvas, and export as JPEG?</p><p>Yes. Yes we can. And it’s faster, more reliable, and doesn’t require loading any external dependencies.</p><p><strong>The insight:</strong> Watching Claude debug in real-time — recognizing the failure, diagnosing the cause, and pivoting to a simpler solution — was more educational than any “How to Process Video in JavaScript” tutorial I could’ve read.</p><p>Then the server crashed. Twice.</p><p>At one point, the page wouldn’t load at all. Server error. We had to clear the Next.js cache, kill processes on ports 3000 and 3001, and restart.</p><p>This is the stuff tutorials don’t show you. The “wait, why is nothing working?” moments. But working through them <em>with</em> Claude (who calmly suggested clearing the cache and restarting) made them feel manageable instead of panic-inducing.</p><h3>🧠 Lesson 3: A Masterclass in Context and Memory</h3><p>Here’s where things got really interesting for me as someone who is dabbling into AI products.</p><p>When building the video analysis feature, Claude had to decide: <em>What information does the AI coach need to do its job well?</em></p><p>This is essentially prompt engineering in action, and I got to watch the decision-making process unfold.</p><p><strong>What gets passed to the AI:</strong></p><p>1. Player description — “Blue jersey, number 7”</p><p>2. Position — Forward, Midfielder, Defender, Goalkeeper</p><p>3. 30 frames extracted from the video at even intervals (more on this journey below)</p><p>4. A carefully crafted system prompt telling Coach Fabian how to analyze footage</p><p><strong>What DOESN’T get passed:</strong></p><p>- The full video file (too large, unnecessary)</p><p>- Previous chat history (this is a separate analysis flow)</p><p>- Complex tactical jargon (the AI is coached to keep it simple for 10–14 year olds)</p><p>The player description part? That actually came from <em>me</em> during our iteration. I realized most game videos have a whole team of kids playing — how would Coach Fabian know which one to focus on? Iterating with Claude helped me see this nuance, and we added jersey color and number inputs to the flow.</p><p>The system prompt itself:</p><p>You are Coach Fabian, analyzing game footage for a youth soccer player (ages 10–14).</p><p>Your approach:</p><p>1. ALWAYS lead with 2–3 specific things they did well</p><p>2. Identify ONE thing to work on (not multiple — don’t overwhelm)</p><p>3. Suggest ONE simple drill they can practice</p><p>Your tone:</p><p>- Encouraging and supportive (“I love how you…” not “You should have…”)</p><p>- Age-appropriate language (avoid complex tactical jargon)</p><p><strong>The insight:</strong> Building AI features isn’t just about the code — it’s about carefully curating what context the AI receives. Too little, and it can’t do its job. Too much, and it gets confused (or expensive). Claude’s approach was surgical: give the AI exactly what it needs to succeed, nothing more.</p><h3>🔬 Lesson 4: Evaluating AI Output is Harder Than Building It</h3><p>Here’s where more real hands-on learning happened. We had a working feature… but was the output actually <em>good</em>?</p><h4>The “Keep Your Head Up” Incident</h4><p>I tested with a clip where the player clearly had his head up the entire play. Coach Fabian’s feedback?</p><p>“Work on getting your head up more when you receive the ball.”</p><p>🤦‍♀️ The exact opposite of what happened.</p><p>This exposed a fundamental limitation: Claude was seeing still frames, not continuous video. You can’t judge head movement from snapshots. It was guessing — and guessing wrong.</p><p>The Fix: Honesty in the Prompt</p><p>We updated the system prompt to be explicit about what it can and can’t observe:</p><p>IMPORTANT CONTEXT: You are seeing 30 still frames, NOT continuous footage. You CAN observe: positioning, spacing, body shape, movement patterns. You CANNOT judge: head movement, reaction time, ball-watching, timing</p><p>DO NOT give advice about things you can’t see in still images.</p><p>The next analysis? Much better. It focused on positioning and movement — things actually visible in the frames.</p><h4>The Edge Case Ping-Pong</h4><p>Then we thought: what about bad inputs? Blurry videos, wrong sport, empty fields?</p><p>Claude added edge case handling to the prompt:</p><p>EDGE CASES — Handle these gracefully:</p><p>- If you can’t identify the player: “I had trouble spotting you…”</p><p>- If video quality is poor: “The video is a bit blurry…”</p><p>- If not soccer footage: “This doesn’t look like soccer…”</p><p>- If no action visible: “I don’t see much action here…”</p><p>Sounds reasonable, right? I tested it with a perfectly clear clip.</p><p>Coach Fabian’s response: I had trouble spotting you in the footage.</p><p>🤦‍♀️ It was <em>too</em> cautious. Given an escape hatch, it took it every time.</p><p>The Fix: Default to Action</p><p>We flipped the framing. Instead of listing reasons to bail out, we made feedback the default:</p><p>IMPORTANT: Default to giving feedback! Even if you’re not 100% sure which player matches the description, pick the most likely candidate and analyze their play. Kids want feedback, not excuses.</p><p>Only use edge case responses when TRULY necessary:</p><p>- Video is completely unwatchable (extremely blurry/dark)</p><p>- Clearly NOT soccer (basketball, swimming, etc.)</p><p>- No players visible at all (empty field, crowd shots only)</p><p>The key insight? Permissive by default, restrictive only when necessary. Same edge cases, totally different behavior.</p><h4>The Frame Count Journey: 6 → 12 → 30</h4><p>We started with 6 frames. Quick math:</p><p>- 30-second video ÷ 6 frames = 1 frame every 5 seconds</p><p>- A lot happens in 5 seconds of soccer 😬</p><p>Claude walked me through the tradeoffs:</p><p>| Frames | Coverage | API Cost |</p><p>| 6 | Miss key moments | ~$0.04 |</p><p>| 12 | Better | ~$0.06 |</p><p>| 30 | Good coverage | ~$0.12 |</p><p>We landed on 30 frames. Still under $0.15 per analysis, but way better coverage.</p><h4>Adding a Safety Net: LLM-as-Judge</h4><p>What if Coach Fabian gives harsh or inappropriate feedback to a 10-year-old? We added a safety check:</p><p>```typescript</p><p><em>// After analysis, run it through Haiku for a vibe check</em></p><p>const safetyCheck = await anthropic.messages.create({</p><p>model: ‘claude-3-haiku’,</p><p>messages: [{</p><p>role: ‘user’,</p><p>content: `Is this feedback appropriate for a 10–14 year old?</p><p>Flag if: harsh, discouraging, or inappropriate.</p><p>“${analysis}”</p><p>Reply: PASS or FAIL`</p><p>}]</p><p>})</p><p>If it fails? Return a safe, generic fallback. The kids never see anything inappropriate.</p><p><strong>The insight:</strong> Building AI features is only half the job. The other half is evaluating outputs and building guardrails. Claude helped me think through evaluation strategies I wouldn’t have considered on my own.</p><h3>🎬 The Final Result</h3><p>Kids can now:</p><p>1. Tap “Analyze My Game” on the home screen</p><p>2. See guidance: “For best results, upload a 15–30 second clip”</p><p>3. Upload a video clip from their game</p><p>4. Describe themselves (“I’m the one in blue, number 10”)</p><p>5. Pick their position</p><p>6. Get encouraging, specific feedback from Coach Fabian (with safety guardrails)</p><p>7. Optionally have Coach Fabian <em>*speak*</em> the feedback aloud</p><p>Under the hood:</p><p>- 30 frames extracted for good coverage</p><p>- Honest prompt that only comments on observable things</p><p>- LLM safety check before showing feedback to kids</p><p>- Graceful handling of edge cases (blurry video, wrong sport, etc.)</p><p>See it in action here: <a href="https://ai-soccer-coach.vercel.app/analyze">https://ai-soccer-coach.vercel.app/analyze</a></p><p>All built in two weekend sessions, with an AI partner who taught me as much as it coded.</p><p>And with cinnamon raisin bread in tow!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uYBDsPDfJYCnUMBl3FbEtA.jpeg" /></figure><h3>💭 Final Thoughts</h3><p>I started this project knowing Claude Code would be collaborative — that’s how coding agents work.</p><p>What I didn’t expect was how much I’d <em>learn</em> from the collaboration.</p><p>It’s more like having a patient, experienced tech lead who:</p><p>- Plans the architecture before diving into code</p><p>- Walks you through the design and asks for your input</p><p>- Pivots gracefully when the first approach doesn’t work</p><p>- Stays calm when things break (which they will)</p><p>- Thinks through evaluation and edge cases with you</p><p>- Teaches you by doing, not lecturing</p><p>Am I a better PM after this weekend? I think so. Not because I could write code but because I <em>understood</em> technical decision making more.</p><p>And honestly? That feels like the future of learning to build.</p><p><strong>Next up</strong><em>: </em>Letting kids specify a key moment (“focus on 0:15–0:25”) for even more targeted feedback. Stay tuned! ⚽</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec7f298086f6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Softest Bread I’ve Ever Made (And I Can’t Stop Making It) ]]></title>
            <link>https://medium.com/@momwhovibecodes/the-softest-bread-ive-ever-made-and-i-can-t-stop-making-it-b536007277df?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/b536007277df</guid>
            <category><![CDATA[baking]]></category>
            <category><![CDATA[cooking]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 08 Feb 2026 19:50:56 GMT</pubDate>
            <atom:updated>2026-02-08T19:50:56.979Z</atom:updated>
            <content:encoded><![CDATA[<p>You know that pillowy, cloud-like bread from Asian bakeries? The kind that’s so soft you almost feel guilty eating it?</p><p>I finally made it at home. And now my family won’t let me stop.</p><p>It’s called Japanese milk bread, and the secret is a technique called <em>tangzhong</em> — basically a cooked flour paste that keeps the bread impossibly soft and fresh for days.</p><p>This recipe is from Joanne Chang’s “<a href="https://www.myersandchang.com/store/pastry-love/">Pastry Love</a>” — specifically her “Alina’s Milk Bread” recipe. If you don’t own this cookbook yet, do yourself a favor. It’s incredible. 📖</p><p>Fair warning: this takes some time. But most of it is waiting around while the dough does its thing. Perfect as a vibe coding companion.</p><h3>What You’ll Need</h3><p>For the Tangzhong (the magic paste)<br>- 1/4 cup (45g) bread flour<br>- 1/2 cup (120g) whole milk<br>- 1/2 cup (120g) water</p><p>For the Bread<br>- All of the tangzhong (above)<br>- 1/2 cup (120g) heavy cream, room temp<br>- 1/2 cup (120g) whole milk, room temp<br>- 2 large eggs, room temp<br>- 2 packets (14g) active dry yeast<br>- 4 3/4 cups (710g) bread flour<br>- 1/2 cup (100g) sugar<br>- 2 teaspoons kosher salt<br>- 1/2 cup (115g / 1 stick) unsalted butter, softened and cut into pieces<br>- 1 egg yolk, for brushing on top</p><p>Makes 2 loaves 🎉</p><h3>Let’s Do This</h3><h4>Step 1: Make the Tangzhong</h4><p>This is the secret weapon.</p><p>Whisk together the flour, milk, and water in a small saucepan. Heat over medium-high, whisking constantly for 6–8 minutes.</p><p>It’ll start out liquidy, then suddenly get thick and paste-like. You want it to leave a trail when you drag the whisk through. Think… pudding consistency.</p><p>Take it off the heat and let it cool to room temp.</p><p>(You can make this a day ahead and keep it in the fridge.)</p><h4>Step 2: Mix the Dough</h4><p>In a stand mixer with the dough hook:</p><p>1. Add the tangzhong, cream, milk, eggs, and yeast<br>2. Mix on low for about 2 minutes until combined<br>3. Add flour, sugar, and salt<br>4. Mix on medium-low for 4–6 minutes until it comes together and feels elastic<br>5. Add butter a few pieces at a time, mixing on medium-low for another 8 minutes</p><p>The dough should be smooth, supple, and a little tacky. Not sticky — just slightly clingy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CmcZwaAQsXwI_tf84uCnaw.jpeg" /></figure><h4>Step 3: First Rise 🕐</h4><p>Cover the bowl with plastic wrap or a clean towel.</p><p>Let it rise at warm room temperature for **1 hour**.</p><h4>Step 4: Punch It Down (Literally)</h4><p>Punch the dough right in the middle. Satisfying, right? 👊</p><p>Fold all the edges into the center, flip it over, cover again, and let it rise for another 30 minutes.</p><h4>Step 5: Shape the Loaves</h4><p>Prep two 9x5-inch loaf pans — butter them generously or use pan spray.</p><p>Flour your work surface, then:</p><p>1. Divide the dough in half<br>2. Divide each half into 4 equal pieces (8 pieces total)<br>3. Roll each piece into an oval, about 5 inches wide and 8 inches long<br>4. Fold the right side to the center, then the left side to meet it<br>5. Roll it up from the bottom like a little jelly roll 🌀<br>6. Place seam-side down in the pan<br>7. Repeat with 3 more pieces so you have 4 rolls nestled side by side<br>8. Do the same with the other 4 pieces in the second pan</p><p>Cover and let rise for 60–70 minutes, until the loaves are puffy and jiggly when you poke them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H_6SaYE5IjvPiSg3-JRRsQ.jpeg" /></figure><h4>Step 6: Bake! 🔥</h4><p>Preheat oven to 350°F.</p><p>Brush the tops with the egg yolk (just whisk it with a fork first).</p><p>Bake for 35–40 minutes, rotating the pans halfway through. You want golden brown tops that sound hollow when you thump them.</p><p>Let them cool in the pans for 15 minutes, then tip them out onto a wire rack.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aulW5pGXH3E0s5yc5Z6cPA.jpeg" /></figure><h3>The Moment of Truth</h3><p>Slice into it while it’s still slightly warm.</p><p>That pull-apart softness? That’s the tangzhong magic. ✨</p><p>Best eaten the same day, but it stays soft in a paper bag for 2 days. Or freeze it (wrapped well) for up to 3 weeks — just toast the slices to freshen them up.</p><h3>The Takeaway</h3><p>This bread takes patience, but zero fancy equipment. Just a stand mixer, some time, and a willingness to let the dough do its thing.</p><p>Happy baking 🍞🍞🍞</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b536007277df" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Gave My AI Soccer Coach a Voice (And a Face) ⚽]]></title>
            <link>https://medium.com/@momwhovibecodes/i-gave-my-ai-soccer-coach-a-voice-and-a-face-361fb1b99680?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/361fb1b99680</guid>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[moms]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 08 Feb 2026 18:57:40 GMT</pubDate>
            <atom:updated>2026-02-08T19:51:59.351Z</atom:updated>
            <content:encoded><![CDATA[<p>It is week two of building. This week, I gave the AI coach a personality upgrade.</p><p>Based on user feedback (thank you kids! 🙌), voice mode was the top requested feature. I get it — what kid wants to <em>type</em> when they’re nervous before a game? They’re pacing around, fidgeting with their cleats, definitely not pulling out a keyboard.</p><p>So here we go.</p><h3>What’s New ✨</h3><p><strong>Coach Fabian</strong> 🏐 The app needed a face. Not a stock photo, not a generic avatar — a character. My kids chose the blob among all the choices that Claude Code came up with. So now there’s a little blob mascot with a soccer ball. He looks excited in Hype Mode (big grin, determined eyebrows 💪) and peaceful in Calm Mode (gentle smile, slow breathing animation 😌). My kids also named it “Coach Fabian”. I’ll take it.</p><p><strong>Voice Mode</strong> 🎙️ Tap the mic, talk to Coach Fabian, and hear him respond. It’s a real conversation now — no typing required. After the coach finishes speaking, the mic automatically turns back on. Just talk. The whole thing flows naturally.</p><p><strong>Mode Themes</strong> 🎨 Hype Mode is now teal and energetic. Calm Mode is purple and peaceful. The whole screen transforms when you switch — background, buttons, message bubbles, everything. It *feels* different, which is kind of the point.</p><p><strong>Fun Touches</strong> ⚡ Floating soccer balls on the landing page. Quick reply chips so kids can tap “I’m nervous” or “Pump me up!” instead of figuring out what to say. Little animations that make the whole thing feel alive.</p><h3>The Voice Thing 🔊</h3><p>Here’s what surprised me: the voice changes everything.</p><p>When my 12 year-old used the text version, he’d type a few words, get a response, then get distracted. Normal kid stuff. 🤷</p><p>With voice mode? He had an actual conversation. Back and forth, like talking to a friend. “I’m really nervous about the game.” *Coach responds.* “But what if I mess up?” *Coach responds.* Three minutes later, he was smiling and ready to go. 😊</p><p>There’s something about hearing a voice that makes it feel more real. More like someone’s actually in your corner.</p><h3>The Tech 🛠️</h3><ul><li>Voice input: Web Speech API (built into browsers, free 🎉)</li><li>Voice output: OpenAI’s text-to-speech API</li><li>The voice: “Fable” — a British accent that sounds friendly and coach-like 🇬🇧</li><li>Still using: Claude API for the AI brain, Next.js, Tailwind, Vercel</li></ul><p>The voice API adds a small cost, but we’re talking cents per conversation.</p><h3>The Process 🍞🎵</h3><p>This time I had Claude Code as my coding partner from the start. I also chose Japanese Milk Bread and Taylor Swift as my main vibe.</p><p>Step 0: Hit play on “Opalite”.</p><p>Step 0.5: Follow <a href="https://medium.com/@momwhovibecodes/the-softest-bread-ive-ever-made-and-i-can-t-stop-making-it-b536007277df">Joanne’s recipe</a>, made the dough, and put it in proof mode in the oven.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CmcZwaAQsXwI_tf84uCnaw.jpeg" /></figure><p>Step 1: Told Claude what I wanted — voice mode, fun UI, mascot character.</p><p>Step 2: Claude wrote a plan. I said “let’s do it.”</p><p>Step 3: Danced to the fun MV a little while Claude built the floating animations and theme system. 💃</p><p>Step 4: Tested voice mode. Realized the mic button was annoying to keep tapping. Asked Claude to make it continuous.</p><p>Step 5: Kid feedback round. “The orange and fire emoji are too much. Let’s name it Fabian and go with a blob”. Claude created Coach Fabian — drew an actual SVG mascot with different expressions for each mode. I did not expect that. 🤯 Switched hype mode background to teal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EB55Xf-9GgTa-f7kTF14TA.jpeg" /></figure><p>Step 5.5: Flipped the dough and proofed for another 30 minutes.</p><p>Step 6: Optimized UI to account for audio lag. There is noticeable lag between each response in voice so I decided to display the text answer while the audio is being prepared.</p><p>Step 6.5: Rolled out the dough and turned them into actual loaves. Continued to proof.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H_6SaYE5IjvPiSg3-JRRsQ.jpeg" /></figure><p>Step 7: Deployed to Vercel 🚀 <a href="https://ai-soccer-coach.vercel.app/">https://ai-soccer-coach.vercel.app</a>. Ta-da, milk bread was done baking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aulW5pGXH3E0s5yc5Z6cPA.jpeg" /></figure><h3>What’s Next 👀</h3><p>IRL:<br>Watch my 12-year old play in his league playoff game 🏆</p><p>Two features left on my original list:</p><ul><li>Video analysis 📹 — upload a game clip, get feedback on what went well and what can be improved</li><li>Personalization 🎯 — the coach remembers their name and position</li></ul><h3>The Takeaway 💡</h3><p>Sometimes a small addition changes everything. Voice turned a chatbot into a coach.</p><p>If you’re vibe coding your own project, don’t underestimate the power of personality. A face, a voice, a color scheme that *feels* right — these things matter. Especially when your users are nervous 10-year-olds.</p><p>Try the new version: <a href="https://ai-soccer-coach.vercel.app/">Coach Fabian</a>✨</p><p>He may be a tad slow in talking but he’s a good listener👂</p><p>Onward and forward to v3 and let’s go, Patriots!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=361fb1b99680" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ciabatta Bread: My Vibe Coding Companion ]]></title>
            <link>https://medium.com/@momwhovibecodes/ciabatta-bread-my-vibe-coding-companion-51f09ca9c9e6?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/51f09ca9c9e6</guid>
            <category><![CDATA[cooking]]></category>
            <category><![CDATA[moms]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Mon, 02 Feb 2026 17:39:28 GMT</pubDate>
            <atom:updated>2026-02-02T17:39:28.651Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Makes 2 Long Loaves | 2-Day Recipe</em></p><p>Folks saw my post about vibe coding while baking and they asked for the recipe 🥖! I am not gate keeping so here you go.</p><p>This recipe comes from the one and only <a href="https://www.myersandchang.com/about/">Joanne Chang</a> and her book <a href="https://www.myersandchang.com/store/pastry-love/"><em>Pastry Love</em></a> — it’s forgiving and fun for all bakers, no matter your experience level.</p><h3>Day 1: Make the Poolish 🌅</h3><p><em>Start 12–16 hours before you want to bake</em></p><h3>Ingredients</h3><ul><li>2 cups / 280g high-gluten bread flour</li><li>1¼ cups / 300g room temperature water</li><li>Pinch of active dry yeast</li></ul><h3>Instructions</h3><p>Combine flour, water, and yeast in a large container. Mix with a wooden spoon or your hands until the flour is totally combined — it will look like a very soupy dough. Cover with a loose-fitting lid or plastic wrap and let sit at room temperature for 12–16 hours.</p><h3>Day 2: Make the Ciabatta 🥖</h3><h3>Ingredients</h3><ul><li>All the poolish from Day 1</li><li>1¼ cups / 300g water, at body temperature (when you stick your finger in, it shouldn’t register as warm or cold)</li><li>1 tablespoon vegetable oil (such as canola)</li><li>½ teaspoon active dry yeast</li><li>3½ cups / 470g high-gluten bread flour</li><li>1 tablespoons kosher salt</li><li>About 2 tablespoons cornmeal, for the baking sheet (optional)</li></ul><h3>Instructions</h3><p><strong>Mix the dough:</strong> In a large bowl, combine water, oil, and yeast. Whisk until yeast dissolves. Add the poolish, bread flour, and salt. Mix with your hands until thoroughly combined — it will seem very soupy and loose. This is the magic of bread: soon this will become lovely, supple dough. Cover with plastic wrap or a lint-free towel and let sit at room temperature for about 40 minutes, until it looks a bit more billowy and airy.</p><p><strong>First fold (40 min):</strong> Take the edges of the dough from the top, bottom, left, and right of the bowl and fold them into the center. Turn the whole dough mass over. Cover and let sit for another 40 minutes. ⏱️</p><p><strong>Second fold (40 min):</strong> Repeat the folding. Cover and let sit for another 40 minutes. In all, the dough will sit for about 2 hours and you’ll fold it onto itself three times. ⏱️</p><p><strong>Shape the loaves:</strong> At this point, the dough should be very loose but soft, silky, and floppy. Lightly flour your work surface and stretch out the dough so it’s about 6 inches wide from side to side and 12 inches long from top to bottom.</p><p>Pick up the top third of the dough and flip it down over onto the middle third. Then take the bottom third of the dough and flip it on top of the middle third. (This is like folding a business letter.) Repeat folding, working in from the shorter ends.</p><p>Turn the whole thing over, then gently pick it up and stretch it over and over a few times until it elongates back to 12 × 6 inches. Use a knife or bench scraper and cut the dough down the middle the long way so you end up with 2 loaves, each 12 × 3 inches.</p><p><strong>Final proof:</strong> Preheat the oven to 450°F and place a rack in the center of the oven. Sprinkle a baking sheet lightly and evenly with the cornmeal. Gently transfer the loaves to the baking sheet, placing them a few inches apart. Lightly cover the loaves with plastic or a lint-free towel and let sit for 30 minutes at warm room temperature to allow the dough to relax and proof a bit.</p><p><strong>Bake:</strong> Bake the ciabatta loaves for 30 to 35 minutes, rotating the baking sheet midway through the baking time, until the loaves are deep golden brown and sound hollow when you thump them in the middle. 🥖✨</p><p><strong>Cool:</strong> Remove from the oven and let cool directly on a wire rack. Slice into quarters for sandwiches.</p><h3>Storage 📦</h3><p>Ciabatta can be stored in a paper bag at room temperature for up to 2 days.</p><p>Here is how mine turned out. Happy baking!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x3f6_taiDgjT79dNAXE8Qg.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=51f09ca9c9e6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Kids Wouldn’t Listen to Me, So I Built Them an AI Coach]]></title>
            <link>https://medium.com/@momwhovibecodes/my-kids-wouldnt-listen-to-me-so-i-built-them-an-ai-coach-671fcadb8636?source=rss-3a46b6e75ae6------2</link>
            <guid isPermaLink="false">https://medium.com/p/671fcadb8636</guid>
            <category><![CDATA[parenting]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[side-hustle]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Mom Who Vibe Codes]]></dc:creator>
            <pubDate>Sun, 01 Feb 2026 20:26:34 GMT</pubDate>
            <atom:updated>2026-02-01T20:54:17.332Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s my first vibe-code project. What should I build that’s fun and useful? I went on a problem-finding journey and landed on an AI soccer coach for my kids.</p><p>Does this sound familiar, soccer moms and dads?</p><p><em>“I don’t want to go to the game — I am going to do badly.”</em> <em>“I’m nervous — the other team is really good.”</em> <em>“I don’t feel like playing.”</em></p><p>As a parent, I’ve tried the usual things: pep talks, breathing exercises, YouTube videos. Sometimes they work. Often they don’t. Kids tune out parents pretty quickly when they’re stressed.</p><p>Let’s try an AI coach instead.</p><h3>What It Does</h3><p>The app has two modes:</p><p><strong>Hype Mode</strong> is for when motivation is low. Maybe they’re tired, or they’d rather stay home and play video games. The AI coach pumps them up — think Ted Lasso energy, but for kids. Short, punchy, encouraging messages that get them excited to play.</p><p><strong>Calm Mode</strong> is for pre-game nerves. Guided breathing, visualization exercises, simple focus techniques. The tone is warm and reassuring. It reminds them that it’s just a game, that they’ve prepared for this, and that they’re going to be okay.</p><p>The whole thing runs in a browser on their phone or tablet. No app to download. No login required. Just open the link and start chatting.</p><h3>Why AI?</h3><p>Here’s the thing about nervous kids: they don’t always want to talk to their parents about it. There’s pressure there, even when we don’t intend it. They know we care about the outcome. They know we’re watching.</p><p>An AI coach has no expectations. It’s not going to be disappointed if they have a bad game. It’s not going to bring it up at dinner. It’s just… there. Ready to help. No judgment.</p><p>I’ve also noticed my kids are more receptive to advice when it doesn’t come from me. A coach, a teacher, even a YouTube video — somehow the same message lands differently. The AI coach fits into that category. It’s external. It’s neutral. And weirdly, that makes it more trustworthy to a 10-year-old.</p><h3>The Tech Stack</h3><p>For anyone curious about how I built it:</p><ul><li><strong>Claude Code</strong> for coding</li><li><strong>Next.js</strong> for the frontend</li><li><strong>Claude API</strong> (Anthropic) for the AI responses</li><li><strong>Tailwind CSS</strong> for styling</li><li><strong>Vercel</strong> for hosting</li></ul><p>Total cost to run: a few dollars a month in API calls.</p><h3>The Process</h3><p><strong>Step 0:</strong> Started a plan in claude.ai — got a full project plan and tech stack. Felt like I have a 24/7 consultant!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7Cbd71SJFkgJ6PyH79MYrg.png" /></figure><p><strong>Step 0.5:</strong> Hit “play” on my playlist to get the vibe going. 🎵🎵🎵</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C8YA_axsNvv1d2VXDwtenA.png" /><figcaption>if you know, you know ;)</figcaption></figure><p><strong>Step 1:</strong> Installed Claude Code on my machine. For those who’ve never interacted with a terminal like me, it was a bit unintuitive — but claude.ai gives you all the bash commands you need.</p><p><strong>Step 2:</strong> Installed Next.js, Tailwind CSS, Supabase for database, and hooked up the Claude API.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZD7dRHO7GRbMb4u5c-abnQ.png" /></figure><p><strong>Step 3:</strong> Set up the project directory with a file describing the project so Claude knows what we’re building.</p><p><strong>Step 4:</strong> Vibe coding session officially commenced. While Claude went to work, I went to make ciabatta. 🍞</p><p><strong>Step 5:</strong> Coding complete! The app was ready for a test drive on my machine locally.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/819/1*3G57mWDYn21JUhB7L4fiPA.png" /></figure><p><strong>Step 6:</strong> First user (my kid) test drove it and provided some useful feedback: <em>“Wow nice! The responses are too long.”</em></p><p><strong>Step 7:</strong> Shared user feedback with Claude, and it went to work updating the prompt.</p><p><strong>Step 8:</strong> Tried it a few more times after the tweak, then deployed to <a href="http://ai-soccer-coach.vercel.app/">Vercel</a>.</p><p>Ciabatta was also baked and ready to eat! Yes, the whole process took about as long as making and baking a loaf of bread. And my killer playlist made it all the better.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x3f6_taiDgjT79dNAXE8Qg.jpeg" /></figure><p>Tested with the 13-year old. He was kind of impressed?!</p><h3>What’s Next</h3><p>This is v1. It works, but there’s a lot I want to add:</p><ul><li><strong>Voice mode</strong> so my kids can just talk to the coach (who types anyways?!)</li><li><strong>Video analysis</strong> where they can upload game clips and get feedback on what they did well and what to work on</li><li><strong>Personalization</strong> so the coach remembers their name, position, and what they’re working on</li></ul><h3>The Takeaway</h3><p>Vibe coding is easy to start and fun to play with. Nothing beats the feeling of seeing your idea come to life and watching your users interact with it in real time.</p><p>Pick an idea, hit play on your favorite playlist, and let’s code.</p><p>One weekend. One working product. One small win.</p><p>Try it out: <a href="https://ai-soccer-coach.vercel.app/">ai-soccer-coach.vercel.app</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=671fcadb8636" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>