<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Adam Schoonmaker</title><description>Senior iOS engineer. I build iOS apps, and the tools that build them.</description><link>https://adamschoonmaker.com/</link><item><title>From zero Garmin experience to shipping our MVP in a month</title><link>https://adamschoonmaker.com/writing/zero-garmin-experience-to-mvp-in-a-month/</link><guid isPermaLink="true">https://adamschoonmaker.com/writing/zero-garmin-experience-to-mvp-in-a-month/</guid><description>I shipped the AllTrails app for Garmin watches to 120 devices this October, on a platform I&apos;d never touched before.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;&lt;img alt=&quot;Two robots labelled agents run a camera rig labelled harness over a row of watches labelled times 120, and a smiling stick figure holds up a photo of a watch. Agents run a rig that photographs the app on 120 devices, and I review the photos instead of the code.&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;2000&quot; height=&quot;1125&quot; src=&quot;https://adamschoonmaker.com/_astro/garmin-photo-rig.BAD3GvUk_7U4Fo.webp&quot;&gt;&lt;/p&gt;
&lt;p&gt;I shipped &lt;a href=&quot;https://apps.garmin.com/apps/bc96ca67-e720-4885-b75d-7c7dd17e55a0&quot;&gt;the AllTrails app for Garmin watches&lt;/a&gt; to 120 devices this October, on a platform I’d never touched before. Agents wrote the code. I reviewed it lightly and focused on verifying outputs to ship with quality.&lt;/p&gt;
&lt;p&gt;This is for engineers using coding agents on a platform they don’t know yet, or for anyone interested in a story of doing that at scale.&lt;/p&gt;
&lt;p&gt;What worked for me was setting the project up for agents from day one: a knowledge base in the repo and a harness that let them check their own work on every device.&lt;/p&gt;
&lt;h2 id=&quot;id-never-touched-a-garmin-but-i-was-intrigued&quot;&gt;I’d never touched a Garmin, but I was intrigued&lt;/h2&gt;
&lt;p&gt;A lot of our users at AllTrails navigate with a Garmin, and getting a route onto the watch took a lot of steps. An app on the watch itself would be a big win. Problem was, nobody at the company knew how to build one.&lt;/p&gt;
&lt;p&gt;Garmin apps run on Connect IQ and are written in Monkey C, a purpose-built language that not many know. I’d never touched Garmin, Connect IQ, or Monkey C. At the time, I didn’t even own a Garmin. But as an avid backpacker, I was intrigued by the phenomenal battery life I could get over my existing Apple Watch Ultra 2, and after a summer of many backpacks I felt motivated to make something happen here.&lt;/p&gt;
&lt;p&gt;Realistically, before agents, it would have taken someone months to ramp up. We had wanted to build this for years and just never had the bandwidth.&lt;/p&gt;
&lt;p&gt;So the weekend after a company off-site, I started long-running agent sessions on one question: could we get anything running at all? Soon, I had a basic prototype of what I envisioned in the simulator, using completely fake data.&lt;/p&gt;
&lt;p&gt;This got me &lt;em&gt;really&lt;/em&gt; excited. I went out and bought a fēnix 8 as a development watch. I then worked with agents to stream dummy data to it from a dev server on my laptop. That brought it to life, and I pitched it &lt;em&gt;hard&lt;/em&gt; across the company. I got leadership buy-in, and it was go time!&lt;/p&gt;
&lt;p&gt;On balance: I wasn’t starting from zero here. &lt;a href=&quot;https://adamschoonmaker.com/projects/feal/&quot;&gt;Feal&lt;/a&gt;, the side-project app I built, had already taught me agent-first habits while I ported the iOS app to Android: snapshot testing, a focus on test quality, and verifying outputs instead of reading every line of code.&lt;/p&gt;
&lt;p&gt;In the AI era, building high levels of trust in hands-off agent output &lt;em&gt;where it’s safe to fail&lt;/em&gt; is worth its weight in gold. That certainly helped me feel confident enough to go for this.&lt;/p&gt;
&lt;h2 id=&quot;write-the-knowledge-down-before-the-code&quot;&gt;Write the knowledge down before the code&lt;/h2&gt;
&lt;p&gt;I knew from the start this had to be an AI-native project, built for agents to work in. So before any serious code, I had agents document everything they could find: best practices for Connect IQ and Monkey C, the platform’s capabilities, and the devices themselves.&lt;/p&gt;
&lt;p&gt;It all went into Markdown files in the repo. The goal was a knowledge base the agents maintain and keep coming back to, so a new session doesn’t redo the research.&lt;/p&gt;
&lt;p&gt;That paid off on the technical parts. Which built-in menus are there? What font sizes does each watch have? How much memory does an app get? What’s the purpose of the app? Anything that needed research became tidily indexed and tucked away for agents.&lt;/p&gt;
&lt;p&gt;With a growing heap of knowledge built into the project, I could brainstorm with an agent and figure out the details with it, instead of me ramping up for ages first.&lt;/p&gt;
&lt;p&gt;The other half of this was the checks. I had the agents fact-check what they wrote against Garmin’s docs, the SDK samples, and the Garmin forums, or even the simulator itself when nothing else had the answer. They also wrote down what they learned, for both themselves and me.&lt;/p&gt;
&lt;h2 id=&quot;prototype-fast-then-rebuild-it-for-agents&quot;&gt;Prototype fast, then rebuild it for agents&lt;/h2&gt;
&lt;p&gt;The prototype was deliberately simple. A server on my laptop had a few dummy endpoints, and I flattened every kind of AllTrails data into a single route type.&lt;/p&gt;
&lt;p&gt;A route answered four questions. How long is this thing? What’s the elevation change? What’s it called? Can I navigate it? For an MVP, that’s really all that mattered, and the platform forced that simplicity anyway.&lt;/p&gt;
&lt;p&gt;Once it got real interest and roadmap time, I rewrote it from scratch. I treated each screen like a finite state machine with explicit states and actions. Since I wasn’t writing the code by hand, I knew that would be key to actually maintaining it.&lt;/p&gt;
&lt;p&gt;That architecture is what made the harness possible. A custom hook reads the running app’s state and can force the app into any state, so an agent can use CLI tools to make a simulator scrub to any screen at any moment.&lt;/p&gt;
&lt;p&gt;Every screen exposes actions that mirror what a user can do, or what the environment can do: a failed network request with a specific error code, or a stubbed response. They run through the same hooks as the real events.&lt;/p&gt;
&lt;p&gt;On top of that came behavior-driven unit tests aiming for complete coverage, then flow tests through the harness on happy and sad paths. By launch, there were nearly a hundred flows.&lt;/p&gt;
&lt;h2 id=&quot;review-the-shape-not-the-lines&quot;&gt;Review the shape, not the lines&lt;/h2&gt;
&lt;p&gt;I understood the high level and lightly reviewed the code, but after reviewing the early work I stopped seriously scrutinizing the logic line by line. It just wasn’t worth the time, nor was it feasible given the roadmap time I had.&lt;/p&gt;
&lt;p&gt;Between prototyping on my own time, chipping away between projects, and the dedicated roadmap time, I put about 4 weeks of “human time” into the build. This was against roughly 5 weeks of “agent time.” As we know, agents can &lt;em&gt;crank out&lt;/em&gt; greenfield code like no other. I couldn’t have &lt;em&gt;possibly&lt;/em&gt; read all of that code in detail.&lt;/p&gt;
&lt;p&gt;Since I &lt;em&gt;really&lt;/em&gt; wanted to make this happen but had limited roadmap time, I chose to adopt a growth mindset and view this as a crucible. Wild timelines can bring wild results, as long as we take care of ourselves along the way, too. The only &lt;em&gt;possible&lt;/em&gt; way to pull off this Garmin dream of mine would be to rethink how I build software. Obviously not by hand, and obviously not “completely vibe-coded” either.&lt;/p&gt;
&lt;p&gt;The sweet spot? Focus on the &lt;em&gt;outputs&lt;/em&gt;, but in a super nerdy, engineering kind of way.&lt;/p&gt;
&lt;p&gt;So I first pointed my energy at building the tooling I needed, starting with that harness I mentioned earlier.&lt;/p&gt;
&lt;p&gt;The harness was a series of command-line tools that could request and read back everything, from the state machine to what’s on screen. From that, agents generated state diagrams that let me check the system and its data flows at a higher level.&lt;/p&gt;
&lt;p&gt;Then came snapshot tests with visual diffs that mirror the flow tests. They let the agents debug and see what they’d built, and they let me review the app one flow at a time. Visual diffing ensured I only reviewed what changed, and unexpected diffs were either proactively fixed by the agent or flagged to me during review.&lt;/p&gt;
&lt;p&gt;By the end, I was mostly reviewing snapshot flows and talking through the bigger questions. What do we need from the server? How should we architect it? Working on it by myself, I was serving as my own product manager and technical project manager.&lt;/p&gt;
&lt;p&gt;Wearing this many hats with nearly complete agency over the app was a blast, but without agents and the strong dev tooling behind it, there’s &lt;em&gt;no way&lt;/em&gt; I’d have shipped this without staying high-level on the coding itself.&lt;/p&gt;
&lt;h2 id=&quot;let-the-harness-see-everything&quot;&gt;Let the harness see everything&lt;/h2&gt;
&lt;p&gt;I did most of my development on that fēnix 8. The harness checked the other 100-plus devices. They come down to about 20 meaningfully distinct screen shapes: circles, small and big rectangles, really short screens where an app can use only a small area at the top left, etc.&lt;/p&gt;
&lt;p&gt;One version of the snapshot tool shows every state on every device, narrowed down by device family. At first, reviewing it all took real time. After that, as long as nothing major changed and nothing got flagged, I was good to go.&lt;/p&gt;
&lt;p&gt;Dozens of times, it caught a layout that just didn’t work on some device.&lt;/p&gt;
&lt;h3 id=&quot;when-german-broke-the-pairing-flow&quot;&gt;When German broke the pairing flow&lt;/h3&gt;
&lt;p&gt;The biggest design changes came from long languages like German, one of the dozen or so languages the app ships in. On the small round watches, the most popular ones like the fēnix, Forerunner, and Venu, the pairing instructions would have scrolled for pages upon pages.&lt;/p&gt;
&lt;p&gt;There was no way to keep the existing design simple and consistent while still fitting a typical watch screen. So I redesigned pairing as a step-by-step flow, one short step per screen, which some technical limits needed anyway.&lt;/p&gt;
&lt;p&gt;The Instinct was the hardest. A round window sits in its top-right corner, cutting off crucial space on an already limited screen. This forced a design where every step got three lengths of text: the ideal description, a medium one, and the shortest (one word).&lt;/p&gt;
&lt;p&gt;Each device then shows the longest one that fits.&lt;/p&gt;
&lt;p&gt;Exposing changing languages in the harness snapshots and flow tests made it possible to iterate and verify that all of this looked reasonable across many devices, in like &lt;em&gt;a day&lt;/em&gt;, instead of a week or two.&lt;/p&gt;
&lt;h3 id=&quot;catching-sneaky-issues&quot;&gt;Catching sneaky issues&lt;/h3&gt;
&lt;p&gt;On top of my manual QA, we also did a bug bash with all the Garmin-owning AllTrails employees we could find. Surprisingly few bugs came up, but we did flag a couple launch blockers where we didn’t like the functionality.&lt;/p&gt;
&lt;p&gt;The fixes for two of our launch-blocker issues compiled on every device and passed every unit test, but introduced sneaky, serious issues.&lt;/p&gt;
&lt;p&gt;The first one was a runtime crash caused by a menu theme that older devices don’t have. It happened on ~50 devices and would have crashed the app on each home screen render.&lt;/p&gt;
&lt;p&gt;The second was a number font with no letters, drawing the pairing code as empty boxes on ~70 devices. This would’ve been a total showstopper.&lt;/p&gt;
&lt;p&gt;Neither of these happened on the typical devices I’d test, of course. Only the snapshot and flow tests caught them, the harness tooling saving the day yet again. Otherwise, these kinds of pre-release fixes on tight timelines could easily have become something much worse.&lt;/p&gt;
&lt;h2 id=&quot;help-agents-help-themselves&quot;&gt;Help agents help themselves&lt;/h2&gt;
&lt;p&gt;The agents kept stumbling on the simulator itself. Out of the box, Garmin’s tooling runs one simulator at a time, and there wasn’t much info out there on running several in parallel.&lt;/p&gt;
&lt;p&gt;I often run several large streams of agent work at once with Robowright (an open-source orchestration system I built for Claude Code), and every workstream needed a simulator. That made the simulators and their verification loops the new bottleneck.&lt;/p&gt;
&lt;p&gt;At first, my dev tooling was a mess. Simulators crashed repeatedly when switching devices, the tooling was flaky, and the agents couldn’t tell whether it was OS-level edge cases or tooling limitations. Some early workarounds were hacky: wait out an observed stall between suites, sometimes for 10-15 minutes at a time.&lt;/p&gt;
&lt;p&gt;It worked, but it wasn’t always trustworthy, and it was &lt;em&gt;way&lt;/em&gt; too slow.&lt;/p&gt;
&lt;p&gt;So I intentionally spent a lot of time on it, even allocating 1-2 of my precious simulator slots. Over a week or two, while other Garmin work streams were in flight, I kept pitching ideas to the agent until it found the crux of the issue. I don’t remember the exact details anymore, but I think it had been mistaking crashes in our own tooling for OS or simulator limits. Once we fixed that, the timeouts and stalls thankfully disappeared.&lt;/p&gt;
&lt;p&gt;I eventually got five simulator slots reliably running at once, with a checkout system so each agent could claim one. A full snapshot run went from over an hour to under thirty minutes, and 2-3 workstreams could happen at once.&lt;/p&gt;
&lt;p&gt;The first time I saw them all on my screen, five simulators rapidly cycling through the app states and devices, I thought: wow, what even is my life anymore? My screen looked insane, but it &lt;em&gt;was&lt;/em&gt; pretty cool.&lt;/p&gt;
&lt;h2 id=&quot;where-being-a-rookie-cost-me&quot;&gt;Where being a rookie cost me&lt;/h2&gt;
&lt;p&gt;Time was the first thing, of course. Even with agents and doing “all the right things,” I learned the hard way every time a layout or flow I’d planned didn’t work on some class of devices.&lt;/p&gt;
&lt;p&gt;The harness itself cost a class of devices. It adds about 12 KB, which pushed the 96 KB devices over the edge: the tooling couldn’t test them, so they’re deferred for now. I bet an experienced Garmin dev would have planned for that, with tricks for app size or per-device builds.&lt;/p&gt;
&lt;p&gt;An on-route map became a spike after launch. With the expertise, I think at least route map visualizations would have made the MVP.&lt;/p&gt;
&lt;p&gt;Sadly, I don’t know what I don’t know, and that’s probably the scariest thing about doing something new like this. I’m confident there are no &lt;em&gt;safety&lt;/em&gt; issues, since the navigation itself is Garmin’s own. But where there is software, there are software bugs, and I’m sure some of the issues we’ll eventually discover are because I’m a total rookie with Garmin.&lt;/p&gt;
&lt;h2 id=&quot;the-repeatable-takeaways&quot;&gt;The repeatable takeaways&lt;/h2&gt;
&lt;p&gt;Next time I’m on a platform or feature I’ve never touched, I’ll build the knowledge base before the code. I’ll have agents fan out far and wide, write it down, and fact-check it. Then I’ll get a rough end-to-end prototype working (a “tracer bullet” of functionality), and re-architect once I know enough. If the code’s in a place I can do it, I’ll focus on building tooling to enable quicker feedback loops for agents.&lt;/p&gt;
&lt;p&gt;A lot of this was enabled by agent skills I’ve built or picked up along the way, too. Some key ones were behavior-driven tests, mutation testing gating, and adversarial review. You can find these skills, and many more, in &lt;a href=&quot;https://adamschoonmaker.com/projects/robowright/&quot;&gt;Robowright&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Drafted from my dictation with an agent, then edited by me.&lt;/em&gt;&lt;/p&gt;
</content:encoded></item><item><title>I built an app that doesn&apos;t want your attention. Can that still work?</title><link>https://adamschoonmaker.com/writing/i-built-an-app-that-doesnt-want-your-attention/</link><guid isPermaLink="true">https://adamschoonmaker.com/writing/i-built-an-app-that-doesnt-want-your-attention/</guid><description>Feal is an emotional journal that won&apos;t ping you unless you ask it to.</description><pubDate>Tue, 03 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;&lt;img alt=&quot;A person labelled you sits on a bench and pats the empty seat beside them for a small, unsure cloud-shaped feeling labelled your emotion, while their muted phone, labelled your phone, leans against the bench leg. Title: Feal: Honor your emotions.&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;2000&quot; height=&quot;1125&quot; src=&quot;https://adamschoonmaker.com/_astro/feal-doesnt-want-your-attention.kMjY_QS6_lt86R.webp&quot;&gt;&lt;/p&gt;
&lt;p&gt;Feal is an emotional journal that won’t ping you unless you ask it to. I built it that way because a kind, loving emotional awareness app should help you notice your feelings, not compete for your attention. An emotional journal works best when it stays quiet, and that’s becoming a rare thing for apps.&lt;/p&gt;
&lt;h2 id=&quot;a-feeling-needs-your-attention-an-app-shouldnt-compete-for-it&quot;&gt;A feeling needs your attention. An app shouldn’t compete for it&lt;/h2&gt;
&lt;p&gt;Very few notifications make it to my lock screen. I let only a handful of apps send them, even though I work in the industry. So many apps are loud and distracting, and I don’t think being inundated like that is good for us. I prefer a pull over a push.&lt;/p&gt;
&lt;p&gt;Years of therapy and reading taught me that emotions are information. They’re there to help, and they need your attention and love. Neglect them, and that’s when they can become a problem. An app built around that idea shouldn’t be one more thing shouting from your phone.&lt;/p&gt;
&lt;p&gt;So the voice I imagined for Feal is a chill, wise therapist you happen to meet somewhere low-key and intimate, over drinks, perhaps. That person doesn’t hassle you, but they care.&lt;/p&gt;
&lt;p&gt;They tell you the simplest version of what you need, give you a slight nudge to check in with yourself, and do it in a laid-back, nonjudgmental, encouraging way. They’re not attached to the results, either.&lt;/p&gt;
&lt;p&gt;That’s why the app has no streaks, charts, or scores. A feeling isn’t a number, and a number quietly suggests something that should be improved. Feal is laser-focused on emotional journaling, and that’s it.&lt;/p&gt;
&lt;p&gt;It’s free today, there are no ads, and I didn’t build it to make money. If it becomes genuinely useful to a thousand people who keep coming back, awesome, &lt;em&gt;mission accomplished&lt;/em&gt;.&lt;/p&gt;
&lt;h2 id=&quot;then-i-forgot-to-open-my-own-app&quot;&gt;Then I forgot to open my own app&lt;/h2&gt;
&lt;p&gt;At first I planned on no reminders or notifications at all. Then I spent half a year building Feal on free nights and weekends between travel, alongside workdays that had turned into managing several agents and workstreams at once.&lt;/p&gt;
&lt;p&gt;The AI thing has been powerful for my work, but it sure hasn’t been good for taking breaks or noticing how I’m doing. One day I realized that even as the person building this emotional journaling app, &lt;em&gt;I was forgetting to use it&lt;/em&gt;. That was the whole point of building it!&lt;/p&gt;
&lt;p&gt;Beta testers shared that sentiment. “I just forget to actually record how I’m feeling,” one texted me. They wanted a reminder, and so did I: one nudge at 10 or 11 in the morning, as the day got going, to set aside a minute if I hadn’t already.&lt;/p&gt;
&lt;p&gt;So Feal has one reminder, and it’s built to be the smallest, most minimal companion it can be. It’s off by default. It sends at most one notification a day, at a time you choose. It skips the days you’ve already logged, and after two weeks away it goes quiet instead of nagging.&lt;/p&gt;
&lt;h2 id=&quot;the-risk-is-getting-lost-in-the-noise&quot;&gt;The risk is getting lost in the noise&lt;/h2&gt;
&lt;p&gt;These days, products typically maximize user growth and habit-building via a zero-sum game for your attention. That game usually involves a barrage of notifications. So many apps do it, it reads as table stakes. Realistically, to succeed big, it is.&lt;/p&gt;
&lt;p&gt;But the great thing about a passion project is getting to define my own success. I wanted a genuinely useful tool for myself, and I wanted to learn from the build while fully owning a public mobile product that lets me use the latest and greatest platform capabilities. I also wanted to experiment with how an app designed to my ideals would do.&lt;/p&gt;
&lt;p&gt;So, the only way Feal brings up reminders on its own is to ask whether you want one, at the moment I decided was right: once you’ve come back a second day and logged an emotion again. It’ll ask at most twice, and never more than once per session.&lt;/p&gt;
&lt;p&gt;I fully acknowledge that design could &lt;em&gt;absolutely&lt;/em&gt; be shooting itself in the foot.&lt;/p&gt;
&lt;p&gt;Some beta testers said they might come back if something reminded them, and they never turned the reminder on. Reminders shipped halfway through beta, so it’s unclear whether everyone had a chance to try them.&lt;/p&gt;
&lt;p&gt;The same goes for the insights that open up from the fifth log and on. It’s been hard to show value quickly without overwhelming someone all at once, and I’m not sure I’ve found the balance yet.&lt;/p&gt;
&lt;p&gt;I know the chill-outside-of-work-therapist persona I dreamed up wouldn’t overwhelm or spam people. But can that actually get traction in our noisy digital lives?&lt;/p&gt;
&lt;h2 id=&quot;where-i-hope-it-helps-people&quot;&gt;Where I hope it helps people&lt;/h2&gt;
&lt;p&gt;What someone should notice after a while with Feal is a higher baseline emotional awareness. Concretely: more decisions that factor in the whole self, broader emotional vocabulary, and deeper emotional relationships with themselves and the people they care about.&lt;/p&gt;
&lt;p&gt;For me, on a typical day, it’s the small things adding up. It’s remembering to take breaks at work, so I’m not exhausted or bummed about where the day went. It’s being able to authentically answer “how are you doing?” when people ask, because I actually know where I’ve been. It’s more about shifting toward thriving than just surviving.&lt;/p&gt;
&lt;p&gt;This app’s approach certainly isn’t for everyone, and it doesn’t try to be. If you need the notification barrage, Feal may be too quiet for you. It’s not for someone brand new to their emotions, and it’s not a stand-in for actual therapy, either.&lt;/p&gt;
&lt;p&gt;For many people, it’ll at least do what one beta tester found: “It gave me 30 seconds to think about just myself. I rarely do that.” That &lt;em&gt;one simple action&lt;/em&gt; can be surprisingly powerful.&lt;/p&gt;
&lt;p&gt;Despite its stance on notifications, Feal does want you to come back, if only because I think it’ll build a positive habit. What it doesn’t want is to commandeer your attention in the meantime.&lt;/p&gt;
&lt;p&gt;If an app built that way seems odd, to me it raises the question: why is it normal that so many apps now demand our attention? A phone is supposed to be a tool &lt;em&gt;for us&lt;/em&gt;, not anyone else. I take that as a reason to consider my own relationship with those apps, and with my phone in general.&lt;/p&gt;
&lt;h2 id=&quot;something-to-try&quot;&gt;Something to try&lt;/h2&gt;
&lt;p&gt;Carve out some time to look at which apps are making it onto your lock screen or into your notifications, and ask what each one is &lt;em&gt;really&lt;/em&gt; doing for you. I’ve turned off push notifications on the majority of my apps and found a more harmonious relationship with my phone; so few things are truly &lt;em&gt;immediately&lt;/em&gt; important.&lt;/p&gt;
&lt;h2 id=&quot;back-to-feal&quot;&gt;Back to Feal&lt;/h2&gt;
&lt;p&gt;If you do want a quiet place to check in emotionally, Feal is on the &lt;a href=&quot;https://apps.apple.com/app/id6775356121?ct=launch-founder-essay&amp;amp;mt=8&quot;&gt;App Store&lt;/a&gt; and &lt;a href=&quot;https://play.google.com/store/apps/details?id=com.adamschoonmaker.feal&amp;amp;referrer=utm_source%3Dfounder-essay%26utm_medium%3Dcontent%26utm_campaign%3Dlaunch-2026-11&quot;&gt;Google Play&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;If you’re in the media and think this is a story worth sharing, the &lt;a href=&quot;https://thefeal.app/press&quot;&gt;press page&lt;/a&gt; has what you’d need.&lt;/p&gt;
&lt;p&gt;More about how I built it as the sole developer is on &lt;a href=&quot;https://adamschoonmaker.com/projects/feal/&quot;&gt;its project page&lt;/a&gt;, and you can read about where the app’s ideas come from at &lt;a href=&quot;https://thefeal.app/philosophy&quot;&gt;thefeal.app/philosophy&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Drafted from my dictation with an agent, then edited by me.&lt;/em&gt;&lt;/p&gt;
</content:encoded></item><item><title>Software engineers are putting down their hammers</title><link>https://adamschoonmaker.com/writing/software-engineers-putting-down-their-hammers/</link><guid isPermaLink="true">https://adamschoonmaker.com/writing/software-engineers-putting-down-their-hammers/</guid><description>Civil and chemical engineers don&apos;t build what they engineer, but software engineers always have, and now agents are taking that part over.</description><pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;&lt;img alt=&quot;Hand sketch: a smiling stick figure labelled “SWE” hands a hammer labelled “code” to two small robots labelled “agents”. Above him, grey arrows fan out from a black dot to the other parts of the job: architect (a blueprint), understand (a tape measure), communicate (a walkie-talkie), verify (a screen with a cross and a check), maintain (a wrench), and a cloud that says “lots more”. A key reads: black dot equals things we’ll still do.&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;2000&quot; height=&quot;1125&quot; src=&quot;https://adamschoonmaker.com/_astro/putting-down-hammers.BNBpfmEO_1QNVlM.webp&quot;&gt;&lt;/p&gt;
&lt;p&gt;Civil and chemical engineers don’t build what they engineer, but software engineers always have, and now agents are taking that part over.&lt;/p&gt;
&lt;p&gt;What’s left of our job is what other engineers have always done. Agents will &lt;em&gt;aid&lt;/em&gt; many of those remaining responsibilities, not replace them.&lt;/p&gt;
&lt;p&gt;This is for software engineers like me who lose sleep over AI-driven job loss. It covers what changed in my own work, why our careers might be okay, and what I’m doing about it.&lt;/p&gt;
&lt;h2 id=&quot;this-idea-started-on-slack&quot;&gt;This idea started on Slack&lt;/h2&gt;
&lt;p&gt;A coworker and I were trading jokes about not having careers much longer. I said I worried about it too. Then I kept typing, and the idea for this post came out in one long message.&lt;/p&gt;
&lt;p&gt;My point was that software has been an odd branch of engineering. Civil engineers design bridges, dams, and buildings, but they aren’t the ones swinging the hammers or laying the concrete. Until recently, by writing the code by hand, we were. I put it this way that day:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;SWE has been a blended role where we’re both the system designers, managers, and the “construction workers.”&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Most of our days went to the construction part, and agents do a lot of it now. I continued, “Now a lot of that ‘construction work’ is agents, but the rest of the role still remains.”&lt;/p&gt;
&lt;p&gt;I stand by that.&lt;/p&gt;
&lt;h2 id=&quot;my-other-engineer-friends-already-dont-swing-the-hammers&quot;&gt;My other engineer friends already don’t “swing the hammers”&lt;/h2&gt;
&lt;p&gt;Two friends I went to engineering school with took the paths I was thinking of that day.&lt;/p&gt;
&lt;p&gt;One studied civil engineering and went into construction. He’s now a project manager at a construction company near Detroit.&lt;/p&gt;
&lt;p&gt;When I got into real estate investing and started having issues with permits, he understood right away. Planning, permits, and building codes are a big part of his job.&lt;/p&gt;
&lt;p&gt;He’s responsible for the higher-level outcomes of construction sites. And while he’s sometimes on-site wearing a safety helmet, he’s never been the one who’s literally swinging the hammer to get the jobs done.&lt;/p&gt;
&lt;p&gt;My other friend studied chemical engineering and became a process engineer at a plant for a household-name consumer goods company.&lt;/p&gt;
&lt;p&gt;His work was designing, monitoring, and troubleshooting the processes that make the product, which for a while meant toothbrushes. School had him expecting a more hands-on job, and over time he learned what the job really was. He’d often be on the assembly line, but he wasn’t ever the one operating it.&lt;/p&gt;
&lt;p&gt;Whether they design the building or run the job site, neither of my friends has been the one actively producing the outputs of their craft. They decide what gets built, decide how things should work, and then they answer for it.&lt;/p&gt;
&lt;p&gt;So if we think of software as a unique field of engineering where we were actively producing the outputs, and now suddenly an alien species known as AI can produce them, it’s natural to panic a bit. Or a lot.&lt;/p&gt;
&lt;p&gt;But the code itself was always just part of the job, and I think we can take comfort in looking to other engineering disciplines as a model for where our field is going.&lt;/p&gt;
&lt;h2 id=&quot;from-100-lines-at-a-time-to-whole-stacks-of-prs&quot;&gt;From 100 lines at a time to whole stacks of PRs&lt;/h2&gt;
&lt;p&gt;This became my reality this summer, while doing a refactor at work.&lt;/p&gt;
&lt;p&gt;Our map screen had a view controller and a view model that were incredibly tightly coupled. Thousands of lines of anti-patterns, spaghetti code, no tests -- I mean, you name it.&lt;/p&gt;
&lt;p&gt;I’d wanted to fix it for three years, and I finally got the go-ahead to spend two weeks on it.&lt;/p&gt;
&lt;p&gt;Earlier this year, when I was chipping away at similarly shaped tech debt with agents, I reviewed their output in 100- to 200-line chunks and steered every step. I’d still catch mistakes.&lt;/p&gt;
&lt;p&gt;This time, AI could work with me on a much more detailed plan: the ideal end state, then about 11 PRs of roughly 300 lines of core logic each. Each PR would include “load-bearing tests.”&lt;/p&gt;
&lt;p&gt;An agent worked through it one PR at a time. The scope roughly doubled once I saw the SwiftUI map view was both massive and acting as a view controller too, and the stack grew to about 15-25 PRs.&lt;/p&gt;
&lt;p&gt;Each PR came back done to spec, close enough to how I would have done it. I fell into a funny loop, planning a couple of PRs ahead and reviewing a couple behind while the stack kept growing. By the end, I was reading whole stacks at once and sending a single round of feedback for all of them.&lt;/p&gt;
&lt;p&gt;I had even stopped needing to read &lt;em&gt;every&lt;/em&gt; line, focusing on the higher-level feedback instead.&lt;/p&gt;
&lt;p&gt;The stack merged the following week. I still caught some minor issues during manual app runs along the way, but I mostly focused on &lt;em&gt;the outputs, the testing strategy, the system design, and the process&lt;/em&gt;.&lt;/p&gt;
&lt;p&gt;I hadn’t written a single line of code by hand, but there was &lt;em&gt;plenty&lt;/em&gt; of work beyond that “hammer swinging” of writing code and reading every single line. Being able to focus on that work was liberating.&lt;/p&gt;
&lt;h2 id=&quot;whats-left-for-software-engineers&quot;&gt;What’s left for software engineers&lt;/h2&gt;
&lt;p&gt;What I checked during that refactor was mostly the tests, aiming for 90% coverage or better, plus the shape of each PR.&lt;/p&gt;
&lt;p&gt;I made decisions like where to carve out a helper object so it could be tested. I decided whether a stop-gap protocol was enough to keep the scope on one subsystem, or whether the fix had to work for the whole app. I also judged whether the separation of concerns and data flows were right.&lt;/p&gt;
&lt;p&gt;Those calls came from ten years in the industry, six years of working at my company, and three years in my area of the product. An agent can read a skill file, but it doesn’t have my gut reaction to a requirement or my sense of where the platform is headed. That judgment is the part of the job I expect to stay ours.&lt;/p&gt;
&lt;p&gt;Trusting agents this much takes an environment built for it, so you start working &lt;em&gt;on the loop&lt;/em&gt; instead of &lt;em&gt;within it&lt;/em&gt;.&lt;/p&gt;
&lt;p&gt;My civil engineering friend has building codes; as software engineers, we often have loose product specs but still write our own constraints. Ours start with unit tests at the seams, then maybe higher-level tests like snapshot, integration, and flow tests. Eventually we build toward agents that test their own work before a human sees it.&lt;/p&gt;
&lt;p&gt;The work is shifting, not evaporating.&lt;/p&gt;
&lt;p&gt;A fair objection is that agents owning judgment is next. They’re already getting better at planning work like my 11-PR plan, and I can’t promise the gap between AI plans and my gut judgment stays forever.&lt;/p&gt;
&lt;p&gt;What I’m counting on, though, is that the context behind a decision, from years in the industry or from working on a specific product, is the last shoe to drop. And hopefully not anytime soon.&lt;/p&gt;
&lt;p&gt;Don’t get me wrong. I &lt;em&gt;loved&lt;/em&gt; the flow state of writing code for hours with Slack turned off, and I do certainly mourn losing it. But I’m also a builder, and I can build &lt;em&gt;far&lt;/em&gt; more now.&lt;/p&gt;
&lt;p&gt;Great software engineering was never just the code. &lt;em&gt;So much&lt;/em&gt; interesting work remains beyond written syntax and markdown plans -- arguably the &lt;em&gt;most&lt;/em&gt; interesting work, and almost certainly the most impactful.&lt;/p&gt;
&lt;h2 id=&quot;but-will-there-be-fewer-of-us&quot;&gt;But will there be fewer of us?&lt;/h2&gt;
&lt;p&gt;The US has about &lt;a href=&quot;https://www.bls.gov/ooh/architecture-and-engineering/chemical-engineers.htm&quot;&gt;22k chemical engineering jobs&lt;/a&gt; and &lt;a href=&quot;https://www.bls.gov/ooh/architecture-and-engineering/civil-engineers.htm&quot;&gt;380k civil engineering jobs&lt;/a&gt;, against about &lt;a href=&quot;https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm&quot;&gt;1.9M Software Developers, Quality Assurance Analysts, and Testers&lt;/a&gt;. If our jobs become more like theirs, does that mean a much smaller profession? Does it become fewer people, managing more work?&lt;/p&gt;
&lt;p&gt;Even from the BLS, the signals are mixed: programmers, a different BLS cut of our industry sized around 110k, are projected &lt;a href=&quot;https://www.bls.gov/ooh/computer-and-information-technology/computer-programmers.htm&quot;&gt;to shrink 7%&lt;/a&gt; over the next decade. Meanwhile, Software Developers, Quality Assurance Analysts, and Testers are projected &lt;a href=&quot;https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm&quot;&gt;to grow 10%&lt;/a&gt;.&lt;/p&gt;
&lt;h3 id=&quot;bet-on-jevons-paradox&quot;&gt;Bet on Jevons paradox&lt;/h3&gt;
&lt;p&gt;&lt;em&gt;Personally&lt;/em&gt;, I’m betting that the demand for software grows faster than agents shrink each software team.&lt;/p&gt;
&lt;p&gt;It’s the Jevons paradox, and I’m already seeing it firsthand.&lt;/p&gt;
&lt;p&gt;The owner of my Krav Maga studio vibe-coded a replacement for the gym software he paid about $200 a month per location for.&lt;/p&gt;
&lt;p&gt;His version is purpose-built for his curriculum, tracks student progression down to specific movements, and adapts the lesson plans for each class to focus on what those specific students need for their next belt test. No way he could buy that, and it’s cheaper than his old software. But he’s a smart guy, and he also told me he expects he’ll need to call “an expert” like me someday when he hits a wall.&lt;/p&gt;
&lt;p&gt;So extrapolating from that concrete example, my logic roughly goes like this:&lt;/p&gt;
&lt;p&gt;Take the expensive, full software teams we’re used to. Let’s say that such a team requires a business opportunity worth $10M+ a year to exist.&lt;/p&gt;
&lt;p&gt;As the costs to build and maintain software drop, and leaner teams can create solutions that would’ve needed a much larger team before, we’ll see that dollar line keep falling.&lt;/p&gt;
&lt;p&gt;First to the single-digit millions, and then maybe even all the way down to the $100k side business that solves a niche problem incredibly well and at a fantastic value. I know my friend with a landscaping business would &lt;em&gt;love&lt;/em&gt; a business management and billing solution dedicated to landscapers if the cost were right.&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Someone&lt;/em&gt; will still have to maintain all of this, too.&lt;/p&gt;
&lt;p&gt;I expect today’s massive software teams to get smaller, with people flowing toward many more small teams. Maybe more solopreneur businesses, or more software agencies dedicated to the vibe-coders.&lt;/p&gt;
&lt;p&gt;I’m betting that adds up to more engineering jobs than we have today, and many new roles we can’t even imagine yet.&lt;/p&gt;
&lt;h3 id=&quot;what-the-layoff-numbers-show&quot;&gt;What the layoff numbers show&lt;/h3&gt;
&lt;p&gt;&lt;img alt=&quot;Hand-drawn bar chart of tech layoffs a year, from Layoffs.fyi: about 165,000 in 2022, 265,000 in 2023, 152,000 in 2024, 123,000 in 2025 and 136,000 through November 1, 2026. The share attributed to AI is shaded: 7% in 2024, 29% in 2025 and 69% in 2026.&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;2000&quot; height=&quot;1125&quot; src=&quot;https://adamschoonmaker.com/_astro/layoffs-by-year.DvInnjFg_2eVUwK.webp&quot;&gt;&lt;/p&gt;
&lt;p&gt;I also suspect that the claims of layoffs driven entirely by AI are overblown. &lt;a href=&quot;https://layoffs.fyi/&quot;&gt;Layoffs.fyi&lt;/a&gt; has some great tracking on this, as seen in the drawing. The number of layoffs has been fairly consistent since the COVID hiring craze, and what’s changed is the explanations for them.&lt;/p&gt;
&lt;p&gt;The way I see this? Follow the incentives.&lt;/p&gt;
&lt;p&gt;A CEO cutting a thousand jobs could give any number of reasonable explanations: the company overhired during COVID, high interest rates remain a business headwind for the foreseeable future, the business simply isn’t performing to expectations and needs to pull back, etc.&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Or,&lt;/em&gt; they could say that a new technology will make the business &lt;em&gt;so much more efficient&lt;/em&gt; that it can achieve the same or better results with fewer expenses (people).&lt;/p&gt;
&lt;p&gt;The &lt;em&gt;outcome&lt;/em&gt; of one thousand layoffs is the same either way, but the &lt;em&gt;narrative&lt;/em&gt; is very different.&lt;/p&gt;
&lt;p&gt;Which narrative would you choose as a CEO? Which company would you rather back as an investor?&lt;/p&gt;
&lt;h2 id=&quot;im-choosing-to-believe-something-good&quot;&gt;I’m choosing to believe something good&lt;/h2&gt;
&lt;p&gt;Nobody &lt;em&gt;really&lt;/em&gt; knows what happens next, me included. &lt;em&gt;But&lt;/em&gt; I’d rather plan on the good outcome than the bad one.&lt;/p&gt;
&lt;p&gt;Hope for the best, prepare for the worst, right?&lt;/p&gt;
&lt;p&gt;Whatever happens, the outcome at an industry level will be the same either way. But if we can stay level-headed during this transition, the journey will be much smoother.&lt;/p&gt;
&lt;p&gt;The folks who come out on the other end of this transformation somewhere successful are going to be the ones who don’t succumb to the doom and gloom, because a person has to believe there &lt;em&gt;can&lt;/em&gt; be a sunny day to build for it.&lt;/p&gt;
&lt;p&gt;So while I’m stashing as much of my income as I reasonably can into savings and investments to prepare for the worst, I’m also doing everything I can to mentally position myself as the person building toward our future.&lt;/p&gt;
&lt;p&gt;If you owned a business or managed a team, which person would you rather have on your team?&lt;/p&gt;
&lt;h2 id=&quot;my-antidote-to-anxiety-is-action&quot;&gt;My antidote to anxiety is action&lt;/h2&gt;
&lt;p&gt;The anxiety, stress, and worry that so many in our industry are feeling today are &lt;em&gt;completely valid&lt;/em&gt;. And boy, do I feel it too.&lt;/p&gt;
&lt;p&gt;What’s helping me is acting on what I control.&lt;/p&gt;
&lt;p&gt;Personal finances? I can control that. I’ve upped my savings rate over the past two years, and I dream of reaching financial independence through frugal-ish living.&lt;/p&gt;
&lt;p&gt;Information diet? I can control that. Every day I decide to skip Reddit doomscrolling and learn from folks actively working out how our industry’s transformation is happening is a win. &lt;a href=&quot;https://newsletter.pragmaticengineer.com/&quot;&gt;The Pragmatic Engineer&lt;/a&gt; is my latest favorite.&lt;/p&gt;
&lt;p&gt;Skills? I can control that, too. I build with agents every working day, embracing them as a mech suit that helps me go further and faster than ever before. I also choose to invest some of my free time outside of work to develop my skills further, and I’ve already seen the dividends at work.&lt;/p&gt;
&lt;p&gt;If you’re losing sleep too, start with &lt;em&gt;one thing&lt;/em&gt; you can control, and let’s get through this crazy time together.&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Drafted from my dictation with an agent, then edited by me.&lt;/em&gt;&lt;/p&gt;
</content:encoded></item></channel></rss>