<?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 Brian G on Medium]]></title>
        <description><![CDATA[Stories by Brian G on Medium]]></description>
        <link>https://medium.com/@brian-g?source=rss-eafd0a8264b7------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*cdBQazUIni4TOEDpfuzLiQ.jpeg</url>
            <title>Stories by Brian G on Medium</title>
            <link>https://medium.com/@brian-g?source=rss-eafd0a8264b7------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 19:06:18 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@brian-g/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[“Engineering” with AI]]></title>
            <link>https://brian-g.medium.com/engineering-with-ai-dcb55e958a1b?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/dcb55e958a1b</guid>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Sun, 27 Sep 2026 20:57:21 GMT</pubDate>
            <atom:updated>2026-09-27T20:57:21.472Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UlnaxrA-rD3Q2-uQ2MtIKQ.png" /><figcaption>Image generated by ChatGPT with the prompt, create an image for a Medium article entitled &#39;Engineering with AI, &#39; and don’t put the title on the image.</figcaption></figure><p>I’m currently in the middle of having Claude, specifically Claude Code, create my cycling dashboard app for me. I’m currently directing the AI to work on software development tasks in the M3 milestone.</p><p>One of the items that I had Claude do for me was to create the project milestones in GitHub.</p><p>Here are all 14 milestones from the CyclometerAI repo:</p><blockquote><strong>M1 — Scaffold &amp; CI</strong> <em>(closed)</em> Xcode project scaffold; TCA + CoreData + SwiftData package setup; CI/CD via GitHub Actions.</blockquote><blockquote><strong>M2 — BLE Clients</strong> <em>(open · 1 open, 9 closed)</em> BLE client — Varia RTL515/RCT715 connect + parse + mock test value; generic HR Profile client (0x180D); CSC Profile client (0x1816).</blockquote><blockquote><strong>M3 — Active Ride Dashboard</strong> <em>(open · 10 open, 8 closed)</em> Active Ride Dashboard — speed, cadence, elapsed time, distance. Widget grid (S05.4 factory default). No radar, no HR yet.</blockquote><blockquote><strong>M4 — Radar &amp; Alerts</strong> <em>(open · 1 open)</em> Radar sidebar visualization (S06); haptic alert escalation L1–L3 via CoreHaptics; three-tone audio alert system (All Clear, Warning, Danger) per Audio.md spec.</blockquote><blockquote><strong>M5 — HealthKit &amp; HR Zones</strong> <em>(open)</em> HealthKit integration (restingHR, maxHR read at ride start); HR zone display with Karvonen formula; BLE HR strap → Apple Watch fallback chain.</blockquote><blockquote><strong>M6 — BLE Speed/Cadence &amp; Wheel Calibration</strong> <em>(open)</em> Wire CSC client (built in M2) into the metrics pipeline; GPS fallback for speed; wheel circumference presets + manual entry + GPS auto-calibration (5% threshold over 500m window).</blockquote><blockquote><strong>M7 — Recording &amp; GPX Export</strong> <em>(open)</em> 1Hz TrackPoint recording to CoreData; vehicle pass event detection; GPX 1.1 export with gpxtpx:TrackPointExtension (HR, cadence, speed) and cyc:VehiclePassEvent waypoints.</blockquote><blockquote><strong>M8 — Navigation &amp; Live Map</strong> <em>(open)</em> Live MapKit map on dashboard (1Hz position, heading-up default); GPX route import from Files app; tribos.studio route integration; turn notifications with audio tone + banner.</blockquote><blockquote><strong>M9 — Ride Summary &amp; History</strong> <em>(open)</em> Post-ride summary screen — distance, time, avg speed, HR zone breakdown, map thumbnail, vehicle pass count, GPX export action; ride history persistence to SwiftData.</blockquote><blockquote><strong>M10 — Settings, Pairing &amp; Onboarding</strong> <em>(open · 2 open)</em> Onboarding flow; device management screen — BLE device list, signal strength, pair/unpair; app settings — units, alert toggles, haptic intensity, Silent Mode override, wheel circumference presets.</blockquote><blockquote><strong>M11 — QA &amp; TestFlight Beta</strong> <em>(open)</em> Full regression QA pass; TestFlight open beta configuration and submission; bug fixes from internal testing; field validation of audio tones and haptic alerts; Garmin BLE program application.</blockquote><blockquote><strong>M12 — App Store Submission</strong> <em>(open)</em> App Store submission — metadata, screenshots, privacy manifest, App Review notes. Phase 1 / MVP complete.</blockquote><blockquote><strong>Phase 2 — Companion, History &amp; Routes</strong> <em>(open · 2 open)</em> Routes tab with list/map views, elevation profile, weather, Strava segments; ride history and detail with map replay and vehicle pass timeline; Apple Watch app + complication; Dynamic Island; dashboard customization; Strava/Garmin Connect export. Target: +2 months post-launch.</blockquote><blockquote><strong>Phase 3 — AR, Power &amp; Platform</strong> <em>(open)</em> Power meter BLE support; ENGO 2 / ActiveLook AR glasses integration with monochrome HUD; segment detection. Target: +4 months post-Phase 2.</blockquote><p>Honestly, when Claude generated this, I didn’t pay too much attention to it. Remember, I’m not really a program manager, but once in a while, I can play one. For M1, I had it generate GitHub issues as Claude saw fit. For M1, I didn’t have it do that; I just had it do it.</p><p>I’m starting to regret not reviewing the milestones with more introspection. I’m thinking of pushing M6 and M10. Right now, I can’t really test the handling of the actual devices because while the Bluetooth code was written in M2, it wasn’t wired completely. When M3 is complete, I’m going to push M6 up in the schedule.</p><p>Another item that is annoying me is the fact that the user profile storage is not due to be worked on until M10, which is nearly the end of the development process, so again, I think I’m going to move that up to either before or after the new order for M6.</p><p>Note: This article has been sitting in my Medium drafts for a long time. Since I wrote about my intentions, I reordered quite a few of the milestones and added a few. I have now completed through M8, including M10, and I’m in the testing and bug fixing phase of M9. I have added M10.5 for fit-and-finish work and M10.6 for WeatherKit integration.</p><h3>My Process</h3><p>This is the basic outline of the process that I’m using with Claude. It is very slow and methodical. Being slow and methodical results in the following:</p><ol><li>It is cost-effective. I’m only paying for the Pro account with Claude, and I’m not paying for extra credits.</li><li>It allows me to verify that each work item works to my satisfaction. This is by far the longest part of the process.</li><li>Allows me to keep the app from drifting too far. I’ve thought about telling Claude to implement an entire milestone of work, but I’m frankly concerned about whether I could verify all of the code, and that means that the quality would drift.</li></ol><p>For M2, I had the AI break down the needed items for the milestone and generate the appropriate items in the project&#39;s GitHub issues. I then asked Claude to generate an implementation plan and asked it to start working on the first task. I used /plan mode and initiated the work with ‘/plan #xx’. Claude works on code development for the issue, creating the unit tests and visual snapshots, and I manually review with SwiftUI previews, the simulator, and running the app on my device. It usually takes about 3 more commits by Claude to get the code to a state that I like. I also use /code-review to have the internal code reviewer look at the code. I also use other LLM engines to review the code that Claude is creating. Codex is extremely helpful when I have the credits.</p><p>I then merge the PR and move to the next work item.</p><p>I’ve been getting more efficient at this process to the point of thinking that I should automate it. Still not sure what will happen at that point.</p><h3>What is the state of the App?</h3><p>I’m sure the question you have is, what is the state of the app? I’m happy with the results so far. It has been a much more complicated app than I originally thought it would be. Frankly, and this is hard to admit, I do not have the skills to write this app.</p><p>The major features that are working:</p><ol><li>Bluetooth sensor integration</li><li>Location tracking</li><li>Import routes from GPX</li><li>Primary dashboard (no customization) that includes a large speed metrics widget, heart rate, heart rate zones, pace, cadence, route directions, and a map.</li><li>Onboarding</li><li>Integration with Apple HealthKit, including initialization of HR zones, zone calculations, and writing to Workouts in the Fitness app.</li></ol><p>Not completed:</p><ol><li>Most of the widgets for the dashboard</li><li>Watch app</li><li>Integration with Strava and other services</li><li>WeatherKit integration</li><li>Dashboard personalization</li><li>Bike tracking</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dcb55e958a1b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My AI told me to]]></title>
            <link>https://brian-g.medium.com/my-ai-told-me-to-1ce4210268cf?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ce4210268cf</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Tue, 26 May 2026 01:30:32 GMT</pubDate>
            <atom:updated>2026-05-26T01:30:32.814Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Db79pcgaEwvBLHDVdE447g.png" /><figcaption>Image for this article generated by Gemini with the prompt: Create a new hero image for a medium article. Do not include text on the image. The article is entitled “My AI told me to”, the focus of the article is about developing an app with AI, thus the cute title.</figcaption></figure><p>I’m about to pull the trigger on having Claude create the initial Xcode project for my app. In doing so, I wanted to make sure that the <a href="https://github.com/brian-g/CyclometerAI/blob/main/assets/PRD.md">Product Requirements Document</a>, the <a href="https://github.com/brian-g/CyclometerAI/blob/main/assets/UX.md">User Experience</a>, the <a href="https://github.com/brian-g/CyclometerAI/blob/main/assets/DataModel.md">Data Model</a>, and more (<a href="https://github.com/brian-g/CyclometerAI/tree/main/assets">https://github.com/brian-g/CyclometerAI/tree/main/assets</a>)</p><h3>Introduction</h3><p>In my first article, <a href="https://medium.com/@brian-g/developing-an-ios-app-with-ai-38baad794c0d">Developing an iOS app with AI</a>, I introduced the app that I’m creating with Claude and talked about some of the UI design options. The focus was on the prompt. However, when I worked with Claude the last couple of days, it was very frustrating because the vision of the dashboard was not clearly communicated in the initial problem.</p><blockquote>The dashboard is the primary UI with the user while riding. It should be a configurable widget oriented interface. Each widget shows a piece of information like speed or cadence. The widget customization interaction should be the similar to the iOS launcher screens, that is to say, press and hold to go into edit mode, drag widgets around, remove widgets, and add widgets. When not in edit mode, tapping a widget should open a bottom drawer showing more info about the data being collected. Each widget has sizes: 1x1, 2x1. The dashboard fits 2x (proportional based on the service dimensions). The dashboard will respond to landscape orientation.</blockquote><p>I recently had a new instance rewrite the vision for the dashboard as:</p><blockquote>The Cyclometer dashboard is not a static screen — it is a rider-configured instrument panel. The design philosophy is borrowed from professional cycling computers and adapted for the iPhone form factor: the rider decides which data matters to them, at what prominence, on how many pages. The app’s role is to make that configuration effortless and to make every widget instantly readable mid-ride, in sunlight, at a glance.</blockquote><h3>Iterative prompts</h3><p>This leads well into my second story, <a href="https://medium.com/@brian-g/prompt-re-prompt-1432d97757d7">Prompt, re-prompt</a>. As you can see, I keep using smaller prompts to generate larger and more detailed prompts. Those deliverables, I then provide feedback to Claude to have additional details flushed out, or Claude will leave a section of open questions that I respond to.</p><h3>Product requirements</h3><p>Next, working on the detailed requirements. You can read the full requirements here: <a href="https://github.com/brian-g/CyclometerAI/blob/main/assets/PRD.md">Product Requirements Document</a>.</p><p>When is an AI hallucination not a hallucination? Or what do you call a desirable hallucination? Anyway, this is where Claude introduced the vision of “safety first”.</p><p>Cyclometer is a safety-first cycling companion for iPhone and Apple Watch that synthesizes real-time sensor data</p><p>I kept it. It isn’t wrong, but it wasn’t my original vision. I also added a better vision for the dashboard, as shown above.</p><h3>Prototyping with AI</h3><p>My next set of articles focused on design and prototyping. <a href="https://medium.com/@brian-g/kill-your-darlings-how-swiftui-dictates-design-2d7f2e005f47?source=user_profile_page---------2-------------eafd0a8264b7----------------------">Killing your darlings</a> focused on how prototyping informed my design (I think for the better), followed by <a href="https://medium.com/@brian-g/prototyping-with-ai-to-learn-950454758348?source=user_profile_page---------1-------------eafd0a8264b7----------------------">Learning SwiftUI</a>, and finally, a discussion on <a href="https://medium.com/@brian-g/design-or-devign-in-the-age-of-ai-322e221b612c">design first or prototype first</a> in the age of AI coding. In summary, both are useful, but can be frustrating for the LLM to keep track of which item is the latest and greatest, which should be in the UX.md.</p><p>Time to generate some code. Wish me luck.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ce4210268cf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Design or Devign in the age of AI]]></title>
            <link>https://brian-g.medium.com/design-or-devign-in-the-age-of-ai-322e221b612c?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/322e221b612c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ui-design]]></category>
            <category><![CDATA[prototyping]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Fri, 22 May 2026 18:58:42 GMT</pubDate>
            <atom:updated>2026-05-22T18:58:42.142Z</atom:updated>
            <content:encoded><![CDATA[<p>I’m a user experience designer with a background in software development. Back before Sketch and Figma, I used Photoshop for creating designs. Sketch and Figma didn’t change the workflow that much. But design is an iterative process.</p><blockquote>I hate doing rework of design files</blockquote><p>Sketch and Figma make iterating on a design much nicer than Photoshop since one can create reusable components. But they are still dead components, and making revisions can be torture.</p><p>There is also the problem of design not aligning with the capabilities of the implementation environment. Some designers prefer not to think about this and instead prefer to think that the developers should just be able to implement anything. However, there is a cost to going completely custom: the time to implement, time to fix issues when the platform upgrades, accessibility, smoothness, and so on. I’ve always believed that a good design helps the user achieve their goals while using the components of the design system and platform as closely as possible.</p><p>These two factors: iterative design and implementation knowledge, lead me to consider using code as a better prototype than using static methods or even the capabilities of Sketch and Figma.</p><p>As I worked through Claude’s request that I design screens for my app, I ran into a problem: Claude wanted a design, but I didn’t want to draw it, as I had a functioning prototype for many of the pages that I developed using LLM integration with Xcode. I consider the prototype to the be source of truth over the design. And, I’m lazy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/603/1*RbjU9St4sAMK6DZPV5VsCQ.png" /><figcaption>Prototype on the left, mockup of the same page and feature on the right. The prototype is more aligned with the iOS 26 design system, and the Sketch mockup is just off in many ways.</figcaption></figure><p>I’m generally happier with prototype rather than the mockup. What I find useful is that the mockup can be used to explore what you want to do without knowing exactly how to implement in a design system like iOS 26.</p><p>The dashboard is very similar, drawing in Sketch to inform a prototype, and the LLM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/877/1*bfxTbvlv2fxAOxrhmLyxSA.png" /><figcaption>Early mockup on the left, dashboard widget mockup, and on the right, the start of a code prototype for the dashboard.</figcaption></figure><p>Once again, the design mockup helps define the vision, inform design decisions (like how to do layout) and then finally a prototype to finalize decisions.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=322e221b612c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Prototyping with AI to Learn]]></title>
            <link>https://brian-g.medium.com/prototyping-with-ai-to-learn-950454758348?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/950454758348</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[swiftui]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Wed, 20 May 2026 19:51:47 GMT</pubDate>
            <atom:updated>2026-05-20T19:51:47.663Z</atom:updated>
            <content:encoded><![CDATA[<p>As I’ve been working through the design decisions in my cycling app, I’ve continued the pattern from my last article: prototyping to understand the capabilities of SwiftUI components better.</p><p>The prototype is actually getting fairly awesome. In some ways, I’m not sure I should really continue building up the UI in Sketch as much as I have.</p><p>Here is what I’ve had Xcode do for me since the last article:</p><ul><li>Added demo data for rides and a past ride viewer</li><li>Added demo data for routes and a router viewer</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rz16KV1T2BVP9XDqq8-6bg.png" /><figcaption>Route list and Ride details</figcaption></figure><ul><li>Added the Setting page</li><li>Added a page (that won’t ship) that illustrates the types of charts available in the Apple chart components</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dM2c26YHPnjulHygygctag.png" /><figcaption>Settings page and Charting gallery page</figcaption></figure><ul><li>Added a custom font for the dashboard</li><li>Started to understand how the layout of the dashboard page will work</li><li>And learn more about SwiftUI</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i2hq6nI-8vN5w2ofUuuQEg.png" /><figcaption>Ride details and Dashboard showing D-DIN Condensed as the typeface for displaying readable numbers on the dashboard.</figcaption></figure><p>One of the features that I didn’t consider initially is the addition of weather to the app. But, I know that as a rider, when planning a ride or a route, I would like to know what the weather is, especially the chance of rain and the direction and speed of the wind during the ride. This prototype helped me understand what I wanted for the app (and I am the primary user.)</p><p>The prompts I have been using are pretty generic. Here is the prompt to build the initial Settings page:</p><blockquote>In Settings do:<br>1. Add 4 sections. The first is unlabeled, the section is called HR Zones, the 3rd is Accounts, and the last is About.<br>2. The first section should have the following preferences: Units (a popup between Imperial and Metric, Wheel size (a picker with standard road bike wheel sizes, Auto-pause (a toggle), Auto-dim (a toggle), “Set Do Not Disturb” (a toggle), and Sensors. Sensors will link to to sensor management page. <br>3. HR Zones should have the 5 standard HR Zones. Each zone should show the current range and have a stepper that allows the user to adjust them up or down. <br>4. In Accounts, there should be an “Add account” button as the last item. Tapping the Add Account button will open an action sheet with the items Strava, Ride with GPS.<br>5. The About section should show the current version, and have a Privacy policy embeded into the app as another navigation page.</blockquote><h3>Learning SwiftUI</h3><p>During this process, I also make tweaks to the pages myself to learn SwiftUI rather than just accept what the LLM wants to do. I found this excellent set of <a href="https://21zerixpm.medium.com/list/100daysswiftuitoexpert-069dcb77c2d2">write-ups on using SwiftUI</a>. I also use Google Gemini to ask questions about SwiftUI, mostly to avoid using all of my Claude tokens.</p><p>SwiftUI is a new mindset. As a web app developer, I disliked React; now I’m embracing the declarative UI definition. I’ve also learned more about the distinct parts of a SwiftUI component, but not enough to even write about here.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=950454758348" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Kill Your Darlings: How SwiftUI dictates design]]></title>
            <link>https://brian-g.medium.com/kill-your-darlings-how-swiftui-dictates-design-2d7f2e005f47?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d7f2e005f47</guid>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[swiftui]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Sat, 25 Apr 2026 16:10:12 GMT</pubDate>
            <atom:updated>2026-04-25T16:10:12.453Z</atom:updated>
            <content:encoded><![CDATA[<p>Getting the design right is tough. I’m currently working through the design of my app. My original model for how the application was going to work was this: An app with 3 bottom tabs: Dashboard, Rides, Settings. The dashboard is the initial page and shows the status of the sensors at the top. When not in a ride, you can long-press to start the customization mode, like is typical in many Apple iOS apps, including the main screen of icons/widgets. My vision looked a bit like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/402/1*BzgS2Ek1LbjunFFe4TaDqw.png" /><figcaption>Intial rough idea for the Cyclometer dashboard.</figcaption></figure><p>One of my goals is to use SwiftUI <strong>and</strong> to follow the <a href="https://developer.apple.com/design/">Apple HIG </a>as much as possible. Looking through the documentation, I wasn’t sure if what I was doing was possible with minimal effort and minimal code. Apple Music was my design pattern of choice. The bottom tabs in Apple Music look like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/312/1*OUrZIktrC4lj-zXN-7WL4Q.png" /><figcaption>Bottom tabs in Apple Music with the <em>Now Playing accessory view</em></figcaption></figure><p>I thought, I can put a <em>Start Ride </em>button in place of the search, right? As I started researching in the SwiftUI documentation about the standard capabilities of the TabView, I soon decided to start prototyping in Xcode to figure out just what the capabilities of the components would be.</p><p>To be honest, I’m still finding my footing with SwiftU. I think the SwiftUI documentation is trash, and I haven’t seen enough examples of good SwiftUI to learn from, so I chose to leverage Xcode’s built-in LLM capabilities using Codex.</p><p>I’m not going to bore you with the entire prompt with Codex (mostly because I lost it), but it was easy to start with the 3 tabs; it was impossible to get the record button put in place. Basically, the TabView does not allow extra buttons to be placed. There is an exception for the search button. The search button, though, is considered a <em>special</em> kind of tab. Back and forth I went with Xcode’s LLM and my manual tweaking of the SwiftUI code. I finally landed on using an accessory view above the TabView that I envisioned would be used to start the recording. Like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/564/1*XQLTajJnocKnRFY33wO7zg.png" /><figcaption>Author’s attempt at using a standard TabView and accessory view.</figcaption></figure><p>I soon started exploring what happens next, namely, start a ride, and then what? Where will the Pause, Finish, and Bell buttons go? I also wanted to reclaim the space for the “Start” and the navigation tabs during a ride. An accessory view can be minimized into the navigation, but only when the main view is scrolled, and the dashboard, by design, is not going to scroll. Dead end.</p><p>I realized that my mental model of the app was starting to fray. This is why one of my personal beliefs is that good design needs to be grounded in the reality of the implementation and the design system. My model of the app was just not congruent with the Apple component system built into SwiftUI. My choice was to change my model or use other components to achieve the same results.</p><p>Along the way, I used both Claude Design and Google Gemini to help me brainstorm additional designs. From a brainstorming perspective, I found Google Gemini’s brainstorming capabilities more helpful than Claude’s. However, neither really met my needs, but both were thought-provoking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1XrhIzTYmAdXZGwJE1rIFA.png" /><figcaption>Google Gemini’s brainstorming output.</figcaption></figure><p>Further noodling on the problem, I went back to the app that I initially chose to model my app after, Apple Music. What struck me about Apple Music, is that I misunderstood, nay overlooked, the Apple Music model, namely, I made the <em>Now playing</em> into a tab, when in reality the song being played is a modal sheet — a modal sheet that when minimized is displayed in an accessory view. Once I discovered that, I directed the Xcode LLM to create an app with 2 tabs: Rides and Settings, a top toolbar with the start button. The start button should display a sheet that confirms the options for the ride as well a place to display sensor status before the ride starts. After the user chooses the “Start Ride” from the modal sheet, they are taken to the Dashboard sheet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/976/1*gViiDcbphrzB4JB1One1Jg.png" /><figcaption>XCode prototype of starting a ride and the interaction to the main Dashboard</figcaption></figure><p>As you can see from the screens above, the interaction is quite a bit different from my initial iteration. Remember, design is about iteration and not falling (too) in love with your babies. More work is needed to refine the interaction and also figure out what the accessory view of an in progress ride looks like as well.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d7f2e005f47" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Producing Product Requirements from AI]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://brian-g.medium.com/producing-product-requirements-from-ai-523667f4a83b?source=rss-eafd0a8264b7------2"><img src="https://cdn-images-1.medium.com/max/1380/1*UWtpPHQ5ad3aKKoBuYCFtQ.png" width="1380"></a></p><p class="medium-feed-snippet">I&#x2019;m going slowly. I&#x2019;m on a journey of discovery, not a journey of profit.</p><p class="medium-feed-link"><a href="https://brian-g.medium.com/producing-product-requirements-from-ai-523667f4a83b?source=rss-eafd0a8264b7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://brian-g.medium.com/producing-product-requirements-from-ai-523667f4a83b?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/523667f4a83b</guid>
            <category><![CDATA[ios-app-development]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Thu, 02 Apr 2026 15:31:39 GMT</pubDate>
            <atom:updated>2026-04-02T15:31:39.452Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Prompt, Re-prompt]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://brian-g.medium.com/prompt-re-prompt-1432d97757d7?source=rss-eafd0a8264b7------2"><img src="https://cdn-images-1.medium.com/max/1376/1*fh6DDWBDbfW0KJQ6J88-XA.png" width="1376"></a></p><p class="medium-feed-snippet">One of the best practices for sucessful prompt writing is to write a prompt and then have the LLM re-write that prompt. When I initially&#x2026;</p><p class="medium-feed-link"><a href="https://brian-g.medium.com/prompt-re-prompt-1432d97757d7?source=rss-eafd0a8264b7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://brian-g.medium.com/prompt-re-prompt-1432d97757d7?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/1432d97757d7</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Fri, 06 Mar 2026 13:43:44 GMT</pubDate>
            <atom:updated>2026-03-06T13:43:44.291Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Developing an iOS app with AI]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://brian-g.medium.com/developing-an-ios-app-with-ai-38baad794c0d?source=rss-eafd0a8264b7------2"><img src="https://cdn-images-1.medium.com/max/600/1*O5LV-UzfmbCBNb2KPOpRjA.png" width="600"></a></p><p class="medium-feed-snippet">I&#x2019;m going to start a series on my process to develop an application with Claude Code for iOS.</p><p class="medium-feed-link"><a href="https://brian-g.medium.com/developing-an-ios-app-with-ai-38baad794c0d?source=rss-eafd0a8264b7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://brian-g.medium.com/developing-an-ios-app-with-ai-38baad794c0d?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/38baad794c0d</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Wed, 04 Mar 2026 22:41:03 GMT</pubDate>
            <atom:updated>2026-03-04T22:41:03.377Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Process of Cancer Identification]]></title>
            <link>https://brian-g.medium.com/the-process-of-cancer-identification-fb65b7f4ee99?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb65b7f4ee99</guid>
            <category><![CDATA[cancer]]></category>
            <category><![CDATA[cancer-survival]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Sat, 12 Feb 2022 01:01:29 GMT</pubDate>
            <atom:updated>2022-02-12T01:01:29.987Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*83uJIZ42Md0vVXR2O4aTEg.jpeg" /><figcaption>CT Scan image of author undergoing a needle biopsy. The needle and the tumor are shown on the right.</figcaption></figure><p><strong>Identification</strong>. The next step in my cancer journey would determine if the lesion in my lung was cancerous, or not. The sampling of the lesion is via CT-assisted needle biopsy. Your imagination of the procedure is probably correct. The medical team sticks a needle in your back, through the lung, and into the tumor using images from a CT scanner to guide the needle to the correct location.</p><p>At this point, I have had 2 CT scans — a low dose calcium score and a full-strength CT scan with contrast.</p><p>This post journals my experience with lung cancer that happened two years ago at the beginning of 2020. Who knew at the time that we’d still be in the middle of a pandemic and that time passes oddly in a global emergency. It hardly seems like yesterday that I was living through these events.</p><p>Between the last CT scan and the needle biopsy felt like an eternity. It was extremely important to <a href="https://brian-g.medium.com/the-struggle-to-remain-emotionally-focused-through-a-cancer-diagnosis-bfa1c16cb921">keep my spirits up and to remind myself that I only knew what I knew</a>. We didn’t know if it was cancer (well pretty sure), and we didn’t know the staging (more on this topic later.) With so many unknowns you can worry yourself to death. Focus on what you know and what you’ll find out.</p><p>The needle biopsy is easy. Nurses give you drugs for the pain and prevent memories of the occasion from forming. You get to relax, if you can call it that, stomach down in a CT scanner. I thought I stayed awake or cogent during the process and while I may have, I don’t remember.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5E9XUmuecTUUnrRgg5mU6g.png" /><figcaption>Progression of the needle into my body</figcaption></figure><p>After waking, I was taken to a recovery room. Needle biopsies could result in a lung collapse, so they keep you under observation and take frequent chest x-rays, to ensure your lung won’t collapse.</p><p><strong>Another wait</strong></p><p>Another wait while the biopsy is accurately identified as cancerous.</p><p><strong>Four days later</strong> around 4:30 PM on a Friday, my phone rang. It was from my doctor’s office. My doctor delivered the news of the biopsy. It was cancer. I had an appointment with an oncologist on Monday.</p><p>After that, I was just in a spin. I called my partner, M, to let her know. As I was, she was devastated. I’m devastated. I have just been given what I think is a death sentence. I weep and cry with M.</p><p>I sat with a mind blank for a while. M was not with me during this time. I can’t remember where she was. Around 7, my phone rang again — it was a private number — but I’m oddly perverse: I answer my phone despite the lack of caller id. It is my doctor. He is calling to check up on me and my mental state. I tell him that I am dealing with it. I am, but I don’t know how. I’m just numb.</p><p>On Monday, the trip to the oncologist is somewhat uneventful. He tells me it is an <strong>adenocarcinoma</strong>. They don’t know much more. Right now, the goal is to identify the stage. Frankly, even at this point, I think I’m pretty early. I have caught lung cancer before symptoms. This knowledge helps feed my positive energy.</p><p>In my next post, I’ll outline the process of the next steps in the diagnosis. My original plan was to publish these articles roughly 2 years after the events. I’m a bit behind as I’m nearly at my 2 year aniversary of my surgery.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb65b7f4ee99" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The struggle to remain emotionally focused through a cancer diagnosis]]></title>
            <link>https://brian-g.medium.com/the-struggle-to-remain-emotionally-focused-through-a-cancer-diagnosis-bfa1c16cb921?source=rss-eafd0a8264b7------2</link>
            <guid isPermaLink="false">https://medium.com/p/bfa1c16cb921</guid>
            <category><![CDATA[depression]]></category>
            <category><![CDATA[cancer]]></category>
            <category><![CDATA[healthcare]]></category>
            <dc:creator><![CDATA[Brian G]]></dc:creator>
            <pubDate>Fri, 26 Nov 2021 15:38:16 GMT</pubDate>
            <atom:updated>2021-11-26T15:38:16.720Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XcpBVN1b-MGT3LZCuirc0Q.jpeg" /><figcaption>Photo by <a href="https://www.pexels.com/@rocktreephoto?utm_content=attributionCopyText&amp;utm_medium=referral&amp;utm_source=pexels">RockTree Photography</a> from <a href="https://www.pexels.com/photo/light-man-people-art-6144393/?utm_content=attributionCopyText&amp;utm_medium=referral&amp;utm_source=pexels">Pexels</a></figcaption></figure><p>“A 2.0 cm cavitating pulmonary nodule or mass with a surrounding thick wall involves the right lower lobe laterally” was the note the radiologist put on my CT scan.</p><p>I previously wrote about how a <a href="https://brian-g.medium.com/how-129-diagnostic-test-saved-me-from-lung-cancer-178e5b5e060f">$139 CT scan found my lung cancer early</a> and saved me from the worst of cancer. I honestly haven’t planned the topics of the articles about my cancer, but it seems that taking care of the emotional side of the diagnosis earlier is best. As a nerd, I immersed myself in looking at the CT scan imagery as a way to distract myself. But that only helps to a certain degree.</p><p>At this point, I kept the diagnosis to just myself and my partner. We didn’t want to needlessly worry others, although I was pretty sure what the prognosis was.</p><p>I had to wait 16 days for the biopsy to confirm if the mass was cancer or not. After the CT scan, my doctor scheduled a needle biopsy. These schedules are the cruelest. There is no end to things that go through your mind when there might be a tumor in your lungs.</p><p>Distressed is the best description of my mental state.</p><h3>I’m going to die.</h3><h3>I’m going to be on oxygen my entire life.</h3><h3>I’ll never be able to ride bicycles again.</h3><h3>I’ll have to get a transplant.</h3><h3>I’ll never really get to live again.</h3><p>My partner had similar feelings. Just say NO to that death spiral of anxiety. Instead of succumbing to anxiety and depression, I made a conscious choice to stay focused on the next step and what I knew at that point. The CT scan indicated <em>something,</em> but that is all we knew.</p><p>It is something, but we don’t know what. I have no symptoms of lung cancer. These are things I told myself. At each step in the process, I continued to reassure myself that there is only so much that I can worry about, only so much that I can do, and more importantly, could not do.</p><p>My destiny was in the healthcare industry’s hands.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bfa1c16cb921" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>