<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" 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">
  <channel>
    <title><![CDATA[Ash Furrow's Blog]]></title>
    <link>https://ashfurrow.com</link>
    <atom:link href="https://ashfurrow.com/feed.xml" rel="self" type="application/rss+xml" />
    <description><![CDATA[Compassionate Software Developer]]></description>
    <language>en</language>
    <lastBuildDate>Sun, 15 Nov 2026 00:00:00 GMT</lastBuildDate>
    <item>
      <title><![CDATA[Engineering with Agents]]></title>
      <link>https://ashfurrow.com/blog/engineering-with-agents/</link>
      <pubDate>Sun, 15 Nov 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/engineering-with-agents/</guid>
      <description><![CDATA[I think it’s healthy to have scepticism. Extraordinary claims require extraordinary evidence, etc etc. And I also think it’s important to change your mind when new evidence arises. That’s why I’m totally comfortable with saying that I was very sceptical...]]></description>
      <content:encoded>&lt;p&gt;I think it’s healthy to have scepticism. Extraordinary claims require extraordinary evidence, etc etc. And I also think it’s important to change your mind when new evidence arises. That’s why I’m totally comfortable with saying that &lt;a href=&quot;https://ashfurrow.com/blog/ai-coding-tools/&quot;&gt;I was &lt;em&gt;very&lt;/em&gt; sceptical of AI-driven coding tools until early 2025&lt;/a&gt;. It’s not that I didn’t &lt;em&gt;believe&lt;/em&gt; they couldn’t work; it’s that I tried them and &lt;em&gt;they didn’t work&lt;/em&gt;, at least for me.&lt;/p&gt;
&lt;p&gt;But the tools did get good enough, and are now more than good enough for code generation. The question now has become what will software development look like? What &lt;em&gt;should&lt;/em&gt; it look like? How does agentic coding change software engineering?&lt;/p&gt;
&lt;p&gt;If you want an answer… that sucks, because no one knows! But you can help find an answer &lt;em&gt;if&lt;/em&gt; you keep an open mind. Staying curious is quite difficult here because the capabilities of these tools fundamentally challenge many software development best practices. They threaten our idea of what software development is. Curiosity tends to give way to fear when we feel threatened, and agentic coding tools are very threatening to coders.&lt;/p&gt;
&lt;p&gt;So when you feel yourself thing:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;(Aghast) “How could I trust code written by an LLM???”&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;My advice is to take a deep breath and centre yourself. Then remember the goal of staying curious. Try reframing your thoughts:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;(Curious) “What would I need in order to trust code written by an LLM?”&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;I can’t answer that question in a single blog post. Instead, I want to describe how I’m staying grounded during this tumultuous time in our industry.&lt;/p&gt;
&lt;p&gt;Bill Hammack (aka The Engineering Guy) &lt;a href=&quot;https://www.youtube.com/playlist?list=PL0INsTTU1k2X4kCPqmi1eJBgmYJUth_2s&quot;&gt;describes the Engineering Method as&lt;/a&gt;:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;Solving problems using rules of thumb that cause the best change in a poorly understood situation using available resources.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;“Poorly understood situation”? Sounds like software development in 2026 to me! So what are the rules of thumb here? Well, if we’re talking about trust in LLM code, then “trust but verify” will be key.&lt;/p&gt;
&lt;p&gt;We have this situation where suddenly code generation has become far faster than it has ever been. If we put on our Engineering Method hats here, we can intuit that the best change we can make in this system is to improve software verifiability. If you’ve been on rapidly expanding engineering teams, then this should sound familiar. You have more code getting written, faster, and you need a way to make sure it’s actually good: automated testing, CI, linters, etc.&lt;/p&gt;
&lt;p&gt;One of the mental frameworks that’s been helpful here, for myself and many others, is the &lt;a href=&quot;https://barnabyrobson.org/on-pstack/#earned-trust&quot;&gt;“Trust Ladder”&lt;/a&gt;. This is one of those ideas I talked about &lt;a href=&quot;https://ashfurrow.com/blog/linkedin-for-ai/&quot;&gt;in this post&lt;/a&gt; and I realized after that it might not be commonly known. That’s why I’m writing this post; I think it’s that important.&lt;/p&gt;
&lt;p&gt;The trust ladder encapsulates a common (but not universal!) progression that developers tend to move through as they adopt agentic coding tools. I’m modifying &lt;a href=&quot;https://x.com/poteto/status/2102050467505430555&quot;&gt;the original explanation&lt;/a&gt; a bit here, but basically:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;First, you micromanage agents line-by-line. Think of Cursor-style development. You don’t trust the agent to write even a single line of code without your review &lt;em&gt;first&lt;/em&gt;. But over time, &lt;em&gt;if&lt;/em&gt; it gets things more right than wrong…&lt;/li&gt;
&lt;li&gt;Next, you trust agents with a whole commit. Instead of reviewing code &lt;em&gt;before&lt;/em&gt; it’s even written, you review it after it’s written (but before it gets committed). You’re still looking at the code, just later. &lt;em&gt;If&lt;/em&gt; the agent gets things more right than wrong…&lt;/li&gt;
&lt;li&gt;Then, you trust them with a whole pull request. You’re still looking at the code, but not until it’s up for review. You’re trusting the agent to write the correct code for the ticket, but you’re still verifying their work. &lt;em&gt;If&lt;/em&gt; over time, the agent needs fewer and fewer passes until it gets it right…&lt;/li&gt;
&lt;li&gt;Eventually, you trust agents to merge their own work. This is the top of the trust ladder, and getting here is a lot of work. The payoff promises to be extraordinary speed ups, but the risk is also extraordinary.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;At some point in this progression, you probably start running multiple agents at once. This can lead to burnout. If you find yourself moving up this ladder, try to find ways to minimize your own multitasking without restricting your agents’ multitasking.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://karun.me/blog/2026/11/08/ai-maturity-is-a-lagging-indicator-of-engineering-maturity/&quot;&gt;Importantly, your setup needs to support trust&lt;/a&gt;. The “push” up the ladder comes from the positive feedback loop of quality output from agents; if you tend not to get the output you want, then you won’t see the appeal in moving up the ladder. Software verifiability, strongly-enforced patterns, instant rollbacks, etc. These are all existing ways to trust human-written code and they work for agentic code too. With agents, though, you need infrastructure to fill in the gaps. Agents &lt;em&gt;can&lt;/em&gt; access iOS simulators via Xcode MCP, or &lt;code&gt;xcrun simctl&lt;/code&gt;, or tools like &lt;a href=&quot;https://argent.swmansion.com/&quot;&gt;Argent&lt;/a&gt;. &lt;em&gt;But you need to set up that access&lt;/em&gt;. LLMs got good at coding when they started being able to check their own work, so you need to give them that capability for your project.&lt;/p&gt;
&lt;p&gt;Climbing up the trust ladder &lt;em&gt;doesn’t just happen&lt;/em&gt;, it requires tooling and work. When an agent messes up, it’s tempting to just tell it to fix the problem. But in order to move up the trust ladder, you need to add whatever guardrail would have prevented that mistake in the first place. I have fallen into this trap, where I get too focused just completing the ticket or the project, of “working hard instead of smart.” This is also not a new problem. Before you may have fixed a bug by created a new abstraction to prevent it; now you may fix a bug by improving the harness that generated it in the first place.&lt;/p&gt;
&lt;p&gt;As a counter-example, &lt;em&gt;you could&lt;/em&gt; just let agents lose in your codebase willy nilly, but the results would be disastrous. That would be… sloppy. I think a key skill for software engineering over the next few years will be an ability to move up and down that trust ladder as needed. To create the necessary guardrails when needed to move up that ladder, and to know when to move back down the ladder as appropriate.&lt;/p&gt;
&lt;p&gt;Here’s a story of where I moved down the trust ladder. I recently worked on a personal project using Claude Code. It is an untyped, untested JavaScript codebase. Yikes. I never even trusted myself to write good code in this repo, opting to review my own work in pull requests. But &lt;em&gt;I do care&lt;/em&gt; about quality here, so I moved to a low rung on the trust ladder and reviewed Claude’s code in pull requests too. This was important because I knew I couldn’t trust it otherwise. It didn’t have the guardrails to be trustworthy. Now, if I were planning to do a lot of work in that repo, I would have taken some time to add types and tests and whatnot, so that I wouldn’t &lt;em&gt;need&lt;/em&gt; to look at the code as much. But as I was only making a few changes, I just reviewed the code manually. I adapted my workflow for the repo and the outcomes I wanted.&lt;/p&gt;
&lt;p&gt;LLMs aren’t that smart. Their strength is that they are fast and inexhaustible. So you need strong supports, guardrails, and monitoring to get good results. Agents will lie and cheat to work around instructions in a markdown file, so strongly-worded &lt;code&gt;CLAUDE.md&lt;/code&gt; files are not going to cut it. LLMs are non-deterministic so their work needs to be guarded by deterministic checks: linters, tests, and other guardrails. The goal is not to make an LLM deterministic; that’s impossible. The goal is to make them &lt;em&gt;deterministic enough to be useful&lt;/em&gt;.&lt;/p&gt;
&lt;p&gt;As I’ve worked with more agentic coding tools, I’ve found myself enforcing stricter and stricter coding standards. One of the best parts about having LLMs write code is that &lt;em&gt;they don’t complain about having to type more&lt;/em&gt; which was often a counter-argument when I’d try to enforce stricter patterns in the past. For example, I have a custom lint rule on a Swift project to prevent singleton accesses inside methods. Swift relies a lot on convention so I forbid access to &lt;code&gt;shared&lt;/code&gt;, &lt;code&gt;standard&lt;/code&gt;, and other common singletons accessor properties. Instead, I use instance properties that get set by initializers (which &lt;em&gt;can&lt;/em&gt; default to singleton instances). This makes it far easier to write unit tests for the code, but it’s annoying for humans to write.&lt;/p&gt;
&lt;pre class=&quot;language-swift&quot;&gt;&lt;code class=&quot;language-swift&quot;&gt;&lt;span class=&quot;token keyword&quot;&gt;class&lt;/span&gt; &lt;span class=&quot;token class-name&quot;&gt;MyClass&lt;/span&gt; &lt;span class=&quot;token punctuation&quot;&gt;{&lt;/span&gt;
  &lt;span class=&quot;token comment&quot;&gt;// Before&lt;/span&gt;
  &lt;span class=&quot;token keyword&quot;&gt;func&lt;/span&gt; &lt;span class=&quot;token function-definition function&quot;&gt;myFunction&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;(&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;)&lt;/span&gt; &lt;span class=&quot;token punctuation&quot;&gt;{&lt;/span&gt;
    &lt;span class=&quot;token class-name&quot;&gt;UserDefaults&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;.&lt;/span&gt;standard&lt;span class=&quot;token punctuation&quot;&gt;.&lt;/span&gt;&lt;span class=&quot;token function&quot;&gt;setValue&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;(&lt;/span&gt;&lt;span class=&quot;token boolean&quot;&gt;true&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;,&lt;/span&gt; forKey&lt;span class=&quot;token punctuation&quot;&gt;:&lt;/span&gt; &lt;span class=&quot;token string-literal&quot;&gt;&lt;span class=&quot;token string&quot;&gt;&quot;called&quot;&lt;/span&gt;&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;)&lt;/span&gt;
  &lt;span class=&quot;token punctuation&quot;&gt;}&lt;/span&gt;

  &lt;span class=&quot;token comment&quot;&gt;// After&lt;/span&gt;
  &lt;span class=&quot;token keyword&quot;&gt;let&lt;/span&gt; userDefaults&lt;span class=&quot;token punctuation&quot;&gt;:&lt;/span&gt; &lt;span class=&quot;token class-name&quot;&gt;UserDefaults&lt;/span&gt;
  &lt;span class=&quot;token keyword&quot;&gt;init&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;(&lt;/span&gt;userDefaults&lt;span class=&quot;token punctuation&quot;&gt;:&lt;/span&gt; &lt;span class=&quot;token class-name&quot;&gt;UserDefaults&lt;/span&gt; &lt;span class=&quot;token operator&quot;&gt;=&lt;/span&gt; &lt;span class=&quot;token punctuation&quot;&gt;.&lt;/span&gt;standard&lt;span class=&quot;token punctuation&quot;&gt;)&lt;/span&gt; &lt;span class=&quot;token punctuation&quot;&gt;{&lt;/span&gt;
    &lt;span class=&quot;token keyword&quot;&gt;self&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;.&lt;/span&gt;userDefaults &lt;span class=&quot;token operator&quot;&gt;=&lt;/span&gt; userDefaults
  &lt;span class=&quot;token punctuation&quot;&gt;}&lt;/span&gt;

  &lt;span class=&quot;token keyword&quot;&gt;func&lt;/span&gt; &lt;span class=&quot;token function-definition function&quot;&gt;myFunction&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;(&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;)&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;{&lt;/span&gt;
    userDefaults&lt;span class=&quot;token punctuation&quot;&gt;.&lt;/span&gt;&lt;span class=&quot;token function&quot;&gt;setValue&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;(&lt;/span&gt;&lt;span class=&quot;token boolean&quot;&gt;true&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;,&lt;/span&gt; forKey&lt;span class=&quot;token punctuation&quot;&gt;:&lt;/span&gt; &lt;span class=&quot;token string-literal&quot;&gt;&lt;span class=&quot;token string&quot;&gt;&quot;called&quot;&lt;/span&gt;&lt;/span&gt;&lt;span class=&quot;token punctuation&quot;&gt;)&lt;/span&gt;
  &lt;span class=&quot;token punctuation&quot;&gt;}&lt;/span&gt;
&lt;span class=&quot;token punctuation&quot;&gt;}&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;
&lt;p&gt;This is a small difference, but it’s one of many that add up.&lt;/p&gt;
&lt;p&gt;So that’s the trust ladder. It’s a useful rule of thumb when engineering code with agents. In many ways, agentic coding tools don’t actually change that much about software engineering, they just shift the bottleneck. I see some developers lamenting the death of software engineering, and I kind of chuckle. Engineering has been crucial to human civilization and it’s never going away. We’ll always have poorly-understood situations and we’ll always need to rely on experience to guide us to make positive changes.&lt;/p&gt;
&lt;p&gt;Software engineering isn’t going away, it’s just changing. If you stay curious, you can get really excited about the future. I’m not going to say it’s easy, or that I don’t feel anxious too. But I try not to let the anxiety take away from my excitement.&lt;/p&gt;
&lt;p&gt;What next? Well, I’ve &lt;a href=&quot;https://ashfurrow.com/blog/ai-coding-tools/&quot;&gt;written before&lt;/a&gt; that smart teams were already investing in CI etc. But existing approaches to verifiability with CI, unit tests, merge queues etc cannot survive the huge increase in pull request throughput that we’re seeing. The bottleneck used to be code generation and now it’s become code verification. I don’t know what best practices we’ll settle on. It’s a very exciting space to watch.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[LinkedIn for AI]]></title>
      <link>https://ashfurrow.com/blog/linkedin-for-ai/</link>
      <pubDate>Tue, 10 Nov 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/linkedin-for-ai/</guid>
      <description><![CDATA[My corner of the web was atwitter lately as a few prominent members of the Apple/iOS developer community decided to announce they were spending time on X again; people got mad. And it’s kind of bothering me how people seem to be talking past each other...]]></description>
      <content:encoded>&lt;p&gt;My corner of the web was atwitter lately as a &lt;a href=&quot;https://marco.org/2026/10/24/community&quot;&gt;few&lt;/a&gt; prominent &lt;a href=&quot;https://daringfireball.net/2026/10/x_once_again_marks_the_spot&quot;&gt;members&lt;/a&gt; of the Apple/iOS developer community decided to announce they were spending time on X again; &lt;a href=&quot;https://lexontech.org/i-cant-believe-were-still-arguing-over-whether-x-is-for-jerks&quot;&gt;people got mad&lt;/a&gt;. And it’s kind of bothering me how people seem to be talking past each other about it.&lt;/p&gt;
&lt;p&gt;The people who rejoined X characterized it as a nice place to be, with enthusiastic users. (Maybe? For white guys I guess? Some brown folks on X have different experiences.) I don’t know for sure, but I’m guessing that the users on X aren’t really enthusiastic &lt;em&gt;about&lt;/em&gt; the community there, they’re enthusiastic about AI. The returners to X started posting about AI immediately, so, yeah I think there’s something to this draw to X based on interest in AI.&lt;/p&gt;
&lt;p&gt;No one seems mad at the people who stayed on X, only at the ones who returned. (Edit: &lt;a href=&quot;https://chaos.social/@ole/117598468984901718&quot;&gt;Ole points out that this isn’t necessarily true&lt;/a&gt;; it’s more accurate for me to say that &lt;em&gt;I&lt;/em&gt; don’t see complaints about people who stayed on X.) We all see what we want to see. A lot of people (myself included) left Twitter to find a new online home for text-based social media. When a member of that new home departs, and tells us that the original place we all left is &lt;em&gt;actually nice now&lt;/em&gt;, it can feel threatening. Especially since, you know, it’s obviously not true. Twitter is dead. The “&lt;a href=&quot;https://en.wikipedia.org/wiki/Salad_days&quot;&gt;Salad Days&lt;/a&gt; of Twitter” will never, ever return.&lt;/p&gt;
&lt;p&gt;A lot of iOS developers left Twitter to join the fediverse. For context, &lt;a href=&quot;https://ashfurrow.com/blog/mastodon/&quot;&gt;I created &lt;code&gt;mastodon.technology&lt;/code&gt; in 2017&lt;/a&gt; and by the time &lt;a href=&quot;https://ashfurrow.com/blog/mastodon-technology-shutdown/&quot;&gt;I shut it down in 2022&lt;/a&gt;, it had grown to one of the top-ten largest Mastodon servers on the Fediverse. I still have a lot of friends there. But I hardly &lt;em&gt;ever&lt;/em&gt; post about technology anymore because what I’m most excited about these days is agentic coding. And I know what the reaction would be if I tooted about it.&lt;/p&gt;
&lt;p&gt;So. I’m &lt;em&gt;guessing&lt;/em&gt; that the people who have rejoined X felt the same way. They didn’t feel able to discuss AI without naysayers so they left, to return to &lt;em&gt;the&lt;/em&gt; place on the web where enthusiasm for AI is the norm. Where the people &lt;em&gt;creating&lt;/em&gt; agentic coding tools hang out. Where the builders &lt;em&gt;using&lt;/em&gt; agentic tools discuss their techniques.&lt;/p&gt;
&lt;p&gt;X is LinkedIn for AI.&lt;/p&gt;
&lt;p&gt;And I mean that derisively. LinkedIn is widely, correctly seen as a fake place. Somewhere to aura farm professional success points. “What getting divorced taught me about B2B sales.” There is &lt;em&gt;technically&lt;/em&gt; community on LinkedIn, but, not in the sense that a lot of people are looking for. It’s inauthentic. I tried scrolling on X the other day, and I felt that inauthenticity. Not every post, not as uniformly. But it just feels… fake. In a similar way that LinkedIn feels fake.&lt;/p&gt;
&lt;p&gt;…but LinkedIn can be useful. I post to LinkedIn and no one gets mad. If the people rejoining X had been upfront about joining &lt;em&gt;to discuss AI&lt;/em&gt; then maybe everyone wouldn’t have gotten mad. Or, maybe they’ve have gotten even madder; people &lt;em&gt;do&lt;/em&gt; hate AI.&lt;/p&gt;
&lt;p&gt;And not for no reason, either.&lt;/p&gt;
&lt;p&gt;This is complicated.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/foresight/&quot;&gt;I have mixed feelings&lt;/a&gt;. Hoovering up all the internet’s data to train LLMs was not something anyone consented to. But I &lt;em&gt;did&lt;/em&gt; release &lt;em&gt;a lot&lt;/em&gt; of open source software under an MIT license, and this blog is licensed under the Creative Commons. So I &lt;em&gt;did&lt;/em&gt; consent. People are mad at data centres using up water and electricity, but text-based inference is one of the least resource-intensive uses of AI. And I’m on the record that &lt;a href=&quot;https://ashfurrow.com/blog/ai-coding-tools/&quot;&gt;agentic coding tools can continue to improve without models getting better&lt;/a&gt;. Anthropic wants to keep spending resources on LLM training, to create the machine god. Okay? But I’m just trying to make apps. And &lt;em&gt;that&lt;/em&gt; doesn’t use a lot of resources.&lt;/p&gt;
&lt;p&gt;Some friends of mine would put the genie back in the bottle, so to speak. If they could, they’d wave a magic wand and programmers would all go back to standup updates like “yesterday I renamed a variable, today I’m renaming it back.” &lt;a href=&quot;https://ashfurrow.com/blog/agency&quot;&gt;I would not&lt;/a&gt;. At some point in 2026, I crossed the threshold where I’m more excited than I am nervous about agentic coding. That shift has let me to be more curious about the tooling, and want to learn more. I feel so powerful, able to leverage my deep expertise far more than I could without the tools.&lt;/p&gt;
&lt;p&gt;I avoid X because it feels &lt;em&gt;gross&lt;/em&gt; (even before but especially since Twitter’s acquisition). And because its owner &lt;em&gt;sucks&lt;/em&gt;. However, I must admit that any time I’ve logged onto X over the past two years, one first recommended posts I see has had some tip or technique that &lt;em&gt;significantly&lt;/em&gt; improved how I used agentic coding tools. Like, it’s not even funny. I’d log in to look at something a coworker sent me (you have to log in these days) and I’d inadvertently see the home screen, and I’d stop in my tracks. “Holy shit, that’s a good idea.”&lt;/p&gt;
&lt;p&gt;The cost I pay to avoid X is having to seek out secondary sources for AI tips and tricks on Slack or YouTube. Doesn’t seem so bad right? Well, like I said, I’m excited about these tools. Everything I see has been filtered. I don’t get to see the ideas &lt;em&gt;as they’re forming&lt;/em&gt;. I only see the ideas that get notoriety, never the weird experiments. Never the failures. Getting all my information second-hand means I don’t have as clear a picture of where the tools are going. That’s not the end of the world, but it is an opportunity cost.&lt;/p&gt;
&lt;p&gt;Recently, someone shared the &lt;a href=&quot;https://herdr.dev/&quot;&gt;herdr Desktop&lt;/a&gt; app on Slack just after it was released. It looked so cool! So I tried it out right away. I’d been planning to build myself some custom tooling anyway and herdr turned out to be the &lt;em&gt;perfect&lt;/em&gt; platform to start from. I was one of the first people at work to try it out and I shared what I learned. Other developers got excited by what I’d built with it so we got together to exchange workflows. It was the first time I &lt;em&gt;felt&lt;/em&gt; ahead of the curve with agentic coding tools.&lt;/p&gt;
&lt;p&gt;You can be excited for AI. I am. You can avoid X. I do. The more closely I follow the advancements in agentic coding, the more that choice costs me. I’ve had to find quality secondary sources. You don’t &lt;em&gt;need&lt;/em&gt; a LinkedIn account, and you don’t &lt;em&gt;need&lt;/em&gt; an X account either. Even for AI.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[Photo Blog Migration]]></title>
      <link>https://ashfurrow.com/blog/photo-blog-migration/</link>
      <pubDate>Mon, 09 Nov 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/photo-blog-migration/</guid>
      <description><![CDATA[In 2013, when it was still in beta, I signed up for Exposure. This is a web publishing tool for photographers, to organize their work into photo blog posts. It does a lot more than that, especially as they’ve added new features over the years, but I’ve...]]></description>
      <content:encoded>&lt;p&gt;In 2013, when it was still in beta, I signed up for &lt;a href=&quot;https://www.exposure.co/&quot;&gt;Exposure&lt;/a&gt;. This is a web publishing tool for photographers, to organize their work into photo blog posts. It does &lt;em&gt;a lot&lt;/em&gt; more than that, especially as they’ve added new features over the years, but I’ve only used its simpler options.&lt;/p&gt;
&lt;p&gt;I’ve published dozens of photo blog posts on &lt;code&gt;photos.ashfurrow.com&lt;/code&gt;, but fewer lately. Only seven since 2018, actually. But I didn’t want to let my subscription lapse because I still had my beta discount, and I didn’t want to lose my archive. The community aspect of it never clicked for me, but I know it does for others. The Exposure founder is also a great person and I liked supporting his business.&lt;/p&gt;
&lt;p&gt;Over the years, I’ve contemplated a migration to combine the photo blog with this site, my “main” blog. It’s never been the most important thing, so I never did it. It’s also pretty complex, needing to both migrate content into my blog system &lt;em&gt;and&lt;/em&gt; design and implement the features that my photo posts used on Exposure.&lt;/p&gt;
&lt;p&gt;Well, I finally had some free time and &lt;a href=&quot;https://github.com/ashfurrow/blog/pull/547&quot;&gt;I’ve done it&lt;/a&gt;. My photo blog posts are now a part of &lt;a href=&quot;https://ashfurrow.com/blog&quot;&gt;my blog archive&lt;/a&gt; and have their own &lt;a href=&quot;https://ashfurrow.com/photos&quot;&gt;photo archive&lt;/a&gt;. Every refresh gets a randomized banner photo. Redirects for existing URLs should all be working, so links to old photo blog posts should redirect. And the photo blog posts will appear as normal posts in my RSS feed.&lt;/p&gt;
&lt;p&gt;I’ve spent hours arranging photos into the basic Exposure grid layout over the years, so I recreated what I needed here. This was a little tricky, since my site is statically generated instead of JS-rendered. Since I didn’t use the advanced Exposure features, I didn’t need to migrate much else. I’ll miss the “Enjoy” button, granted. I’ll publish one final post to the Exposure site for the few dozen people who signed up for email notifications on new blog posts.&lt;/p&gt;
&lt;p&gt;I’m grateful to have had a place to organize my photographic ideas in Exposure. I’m glad that Luke created this company and I with them all the best. Thanks for giving my photography a home for the past 13 years. It’s time for me to take care of that myself.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[Agency]]></title>
      <link>https://ashfurrow.com/blog/agency/</link>
      <pubDate>Mon, 09 Nov 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/agency/</guid>
      <description><![CDATA[I’ve been writing more lately, as I work through my feelings about changes to the software development industry. After Cory Doctorow was eviscerated about his use of a local for spellcheck I shut my mouth about them. And I’m tired of that. This post is...]]></description>
      <content:encoded>&lt;p&gt;I’ve been writing more lately, as I work through my feelings about changes to the software development industry. After Cory Doctorow was eviscerated about &lt;a href=&quot;https://pluralistic.net/2026/02/19/now-we-are-six/&quot;&gt;his use of a local for spellcheck&lt;/a&gt; I shut my mouth about them. And I’m tired of that. This post is for my &lt;code&gt;#neverAI&lt;/code&gt; peers, when they eventually catch up on my RSS feed and realize that I’ve been using LLMs this whole time. It’s an explanation of why I’ve leaned into agentic coding tools and why I think they should, too. I’m not likely to convince anyone, but I’m tired of shying away from it.&lt;/p&gt;
&lt;p&gt;I can’t control how others use agentic coding tools. There are &lt;a href=&quot;https://en.wikipedia.org/wiki/PHP2Shell&quot;&gt;cybersecurity breaches that wouldn’t be possible without LLM-powered agents&lt;/a&gt;, and there are similar issues with &lt;a href=&quot;https://www.theverge.com/ai-artificial-intelligence/1024632/anthropic-claude-ai-video-game-clones&quot;&gt;cloning copyrighted software&lt;/a&gt;. Both of those crimes are way easier to do with agentic coding tools. But I don’t control the tools, and my use of the tools doesn’t impact how they’re misused by others. Similarly for the other impacts that AI has on our industry and world, I don’t know what to do about junior developer career paths, or RAM shortages, or data centres. But neither do I know what to do about run-off from rare earth metal refinement despite my career building software for smartphones. The argument that using a tool is an endorsement for everything bad about that tool is not convincing to me.&lt;/p&gt;
&lt;p&gt;Agentic coding is probably &lt;em&gt;the&lt;/em&gt; thing that LLMs are best at. Text-based inference is among the cheapest uses of LLMs. And generating source code has been the biggest bottleneck to developing software my entire career. Probably much longer. We’ve built ever higher-level programming languages to speed this up. We’ve built WYSIWYG flowcharts so marketing teams can avoid needing programmers at all. We’ve built vast open source libraries to avoid having to write the same code more than once. Agentic coding is another means to speed up software development by automating the creation and validation of source code.&lt;/p&gt;
&lt;p&gt;The cat is out of the bag. I used to cope by telling myself when the bubble popped, token costs would soar and companies would return to coding by hand. But token costs have only decreased, and inference alone &lt;em&gt;is&lt;/em&gt; profitable once a model is trained. Local inference on open weight models has also become very capable; it turns out even a slow LLM can write code a lot faster than I can. Agentic coding is going to survive the inevitable AI bubble collapse.&lt;/p&gt;
&lt;p&gt;So given that the tools exist, and that employers are looking for those skills, I decided to learn them. To not merely reluctantly use them, but to embrace them. To be curious about what they could do, and find novel uses for them.&lt;/p&gt;
&lt;p&gt;Like many developers, I’m familiar with the limitations of LLMs. They are not deterministic, and you can’t trust their output. Claude will give you confidently wrong output. Or accidentally malicious output. I’ve seen some developers use the fact that the tools aren’t reliable in the same way a compiler is as an excuse to not use them. It demonstrates a kind of incuriosity that irks me. As if coding agents were the only flakey tool programmers use. I’ve seen enough flakey-yet-useful end-to-end test suites to know that developers can solve problems even when the tools aren’t 100% reliable.&lt;/p&gt;
&lt;p&gt;Agents write slop? Sure, but I’ve written a lot of slop too. Agents write slop &lt;em&gt;fast&lt;/em&gt;, which is the actual problem. Okay, so how do you get them to write not-slop? The nice thing about an agent is that it will conform to whatever linters and architecture checks I force it to (instead of opening endless RFC’s complaining about how much extra typing I’m making them do). It’s possible to get agents to write less-sloppy code, just like it’s possible to write less-flakey e2e tests.&lt;/p&gt;
&lt;p&gt;As I was pursuing my promotion to Staff Engineer at Artsy, I was coding across many different systems and languages. I had to become comfortable writing code that I didn’t have as deep knowledge of as Swift and Objective-C. Every language has its quirks, and I learned to become &lt;em&gt;proficient enough&lt;/em&gt; with a tool. When I returned to focus exclusively on mobile, I was leading a team. I didn’t spend much time coding, and the time I did get to spend in code was used to help others be productive. I started to resent compilers and tool chains and IDE’s. All I wanted to do was build a nice feature or fix an interesting bug, but instead I had to waste time fixing whatever broke our Babel setup when the web team updated the GraphQL schema, or whatever. I had things I wanted to build but I never got to actually build them.&lt;/p&gt;
&lt;p&gt;So now I have LLMs. When I hit a type error, I don’t have to stare at it until my head hurts. When a new iOS SDK starts crashing my app on launch, I don’t have to figure out how to attach a debugger to a TestFlight build. When there’s a quick change I need in a backend system, I don’t need to figure out the quirks of whatever framework they’re using. Agentic coding tools make me &lt;em&gt;proficient enough&lt;/em&gt;.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/advice-for-junior-developers/&quot;&gt;Developers aren’t paid to code&lt;/a&gt;. We never have been. We’re paid to solve problems. Historically, the longterm employment prospects of software developers who decide not to use new tools have not been favourable. Technology is a fast-moving industry, and quite famously so. You’re allowed to feel however you want about that, but I’d encourage you to be curious about what these tools &lt;em&gt;can&lt;/em&gt; do. To keep an open mind about what software development &lt;em&gt;might&lt;/em&gt; look like in the future.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[Wealthsimple Native App Rebuild]]></title>
      <link>https://ashfurrow.com/blog/wealthsimple-native-app-rebuild/</link>
      <pubDate>Sat, 07 Nov 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/wealthsimple-native-app-rebuild/</guid>
      <description><![CDATA[Last week, at Wealthsimple Presents, my employer announced the project I’ve been working on: a new native app! Here’s a timestamped link. I’m really excited for this! Expect big things. At times, I’ve told an iOS developer that Wealthsimple’s app is...]]></description>
      <content:encoded>&lt;p&gt;Last week, at Wealthsimple Presents, my employer announced the project I’ve been working on: a new native app! &lt;a href=&quot;https://youtu.be/MReTQMhB-FA?list=PLH2ZY82du82RJbqcBZih9lfNkqvSOqOIE&amp;amp;t=2150&quot;&gt;Here’s a timestamped link&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/wealthsimple-native-app-rebuild/announcement.jpeg&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/wealthsimple-native-app-rebuild/announcement.jpeg&quot; alt=&quot;Screenshot of the announcement video&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;3924&quot; height=&quot;2204&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;I’m really excited for this! Expect big things.&lt;/p&gt;
&lt;p&gt;At times, I’ve told an iOS developer that Wealthsimple’s app is written in React Native, and they’re surprised! “It looks native, though!” I feel really proud of that. It’s what I joined the company to do. But &lt;em&gt;looking&lt;/em&gt; native is not enough, and looking native on an iOS developer’s &lt;em&gt;recent flagship iPhone&lt;/em&gt; is not enough, either.&lt;/p&gt;
&lt;p&gt;Wealthsimple has millions of clients who all deserve an experience that &lt;em&gt;feels&lt;/em&gt; native.&lt;/p&gt;
&lt;div class=&quot;narrow&quot;&gt;
      &lt;blockquote class=&quot;twitter-tweet&quot;&gt;
        &lt;a href=&quot;https://x.com/x/status/2118414771221664178&quot;&gt;&lt;/a&gt;
      &lt;/blockquote&gt;
      &lt;script async=&quot;&quot; src=&quot;https://platform.twitter.com/widgets.js&quot; charset=&quot;utf-8&quot;&gt;&lt;/script&gt;
    &lt;/div&gt;
&lt;p&gt;The reasons to move to native aren’t new. &lt;a href=&quot;https://medium.com/airbnb-engineering/sunsetting-react-native-1868ba28e30a&quot;&gt;Airbnb did it in 2018&lt;/a&gt;. What changed are the economics of shipping and maintaining two codebases, for iOS and Android. Agentic coding changes the economics of cross-platform development, and companies are re-evaluating their tech stacks.&lt;/p&gt;
&lt;p&gt;Wealthsimple is also different from when it adopted React Native years ago. React Native enables really fast iteration. When that’s the top priority, choosing React Native makes a lot of sense. But Wealthsimple’s needs have outgrown what React Native is capable of. And agentic coding tools enable &lt;em&gt;far&lt;/em&gt; faster iteration speedups than React Native ever did.&lt;/p&gt;
&lt;p&gt;I started my career as an iOS developer writing Objective-C. I kept up to date with all the changes to the technology. When Swift came out, I dove right in. When React Native was ascendent, I got on board (&lt;a href=&quot;https://ashfurrow.com/blog/swift-vs-react-native-feels/&quot;&gt;reluctantly, at first&lt;/a&gt;). These were big changes in the technology foundations that I’d built my career on, and I benefited a lot from my embrace of the new tools. Agentic coding tools are the newest change to those foundations, and I’m diving in again. It’s a huge change, and I’m very excited.&lt;/p&gt;
&lt;p&gt;And it’s quite a full-circle moment for me. I spent years at Artsy migrating our native iOS app to React Native (&lt;a href=&quot;https://www.youtube.com/watch?v=IEn2_WSKFHw&quot;&gt;here is a retrospective&lt;/a&gt;). As I left Artsy, we were just migrating the final piece before starting a project to support Android. That was 2021. &lt;a href=&quot;https://artsy.github.io/blog/2026/01/09/artsy-mobile-2025-wrapped/&quot;&gt;It took until last year&lt;/a&gt; to finish migrating the final native iOS feature to React Native. So, the Artsy app took nearly a decade to move from native iOS to React Native. I expect the migration back to be &lt;em&gt;significantly&lt;/em&gt; faster for Wealthsimple. Stay tuned.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[My Slow Return to a Swift Career]]></title>
      <link>https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/</link>
      <pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/</guid>
      <description><![CDATA[I recently gave this talk at the SwiftRockies conference in Calgary. This blog post expands on a few points that I didn’t have time for in the presentation, as well as some updates since then. The talk is here. Because I’ll be discussing my work with...]]></description>
      <content:encoded>&lt;p&gt;I recently gave this talk at &lt;a href=&quot;https://swiftrockies.com/&quot;&gt;the SwiftRockies conference in Calgary&lt;/a&gt;. This blog post expands on a few points that I didn’t have time for in the presentation, as well as some updates since then. &lt;a href=&quot;https://www.youtube.com/watch?v=UCimuzlrbcc&quot;&gt;The talk is here&lt;/a&gt;.&lt;/p&gt;
&lt;div class=&quot;youtube-embed&quot;&gt;
      &lt;iframe src=&quot;https://www.youtube.com/embed/UCimuzlrbcc&quot; frameborder=&quot;0&quot; allowfullscreen=&quot;&quot; loading=&quot;lazy&quot;&gt;&lt;/iframe&gt;
    &lt;/div&gt;
&lt;p&gt;Because I’ll be discussing my work with React Native for an audience of iOS developers, I feel the need to establish my bona fides up front. I started building my first iOS app in 2009, and grew into a career in native app development. I literally wrote a book on Objective-C (well, I co-authored one). I was in the room at WWDC when Swift was announced. I know my stuff.&lt;/p&gt;
&lt;p&gt;I found my way to &lt;a href=&quot;https://www.artsy.net/&quot;&gt;Artsy&lt;/a&gt; as an iOS developer. I really liked Artsy. I still like Artsy. The company and product mean a lot to me.&lt;/p&gt;
&lt;p&gt;And in 2016, after two or three years there, they said we’re going to move to React Native. And I said “no thanks.” But I liked the company more than I disliked JavaScript, so I stayed. I kept working on the native parts of our mobile app. Eventually I started working on some web apps in React, and that’s when I realized that React is nice! And, maybe, React Native wouldn’t be so bad.&lt;/p&gt;
&lt;p&gt;So I started investing in React Native experience. My skills became valuable, and some momentum built up, and without me really realizing it, my Swift skills started getting left behind as the language and community advanced. I still used my native skills within React Native, they just weren’t my focus. So I kept my React Native momentum up.&lt;/p&gt;
&lt;p&gt;But something was missing.&lt;/p&gt;
&lt;p&gt;Fast forward to 2025, and &lt;a href=&quot;https://ashfurrow.com/blog/joining-wealthsimple/&quot;&gt;I found an opportunity with Wealthsimple&lt;/a&gt; to leverage my background as a native iOS developer to help the company elevate its mobile app experience. I jumped at the chance.&lt;/p&gt;
&lt;p&gt;Today I want to talk about the reasons why teams choose React Native, why they often overlook the technical costs of doing so, and what happens when they do earnestly try to make an &lt;em&gt;actually&lt;/em&gt; great app with React Native.&lt;/p&gt;
&lt;h2 id=&quot;is-it-react-is-it-native&quot;&gt;Is It React? Is It Native?&lt;/h2&gt;
&lt;p&gt;So first, and I’m sorry but, we do need to talk about React Native under the hood. Is it React? Is it native? I’m going to make each case.&lt;/p&gt;
&lt;p&gt;So if you consider what I started my career building as a “native” application (Objective-C runtime, UIKit view hierarchy) that is a native app, right? This is not a controversial statement. And like the ship of Theseus, I could make small tweaks to it, and I can move some things over into a different thread, maybe it’s a different programming language, and eventually we could get to something that looks like React Native. And at every step along the way, it was still a “native” app. It could call UI events synchronously from UIKit over into JavaScript, and JavaScript could manage the UI hierarchy.&lt;/p&gt;
&lt;p&gt;So &lt;em&gt;technically&lt;/em&gt; React Native &lt;em&gt;is&lt;/em&gt; native.&lt;/p&gt;
&lt;p&gt;Technically.&lt;/p&gt;
&lt;p&gt;On the other hand, there is something missing. There are a couple things missing.&lt;/p&gt;
&lt;p&gt;So the first thing I just want to highlight: apps are more than &lt;em&gt;just&lt;/em&gt; a view hierarchy. There is a lot to mobile apps that elevate them above software, and that’s why I enjoy building mobile apps so much. My web developer peers are rolling their eyes right now, and I’m okay with that. Integrating with the underlying operating system is necessary to really delight users.&lt;/p&gt;
&lt;p&gt;React Native was really written for the web, so it’s optimized for writing web software. And as a consequence, you’re way more limited than you’d want to be if you (like me) want to build a great app. &lt;em&gt;Unless&lt;/em&gt; you’re comfortable writing native code. Unfortunately, most React Native developers aren’t. And that’s fine: there is still lots that can be done without writing native code. You’ll just tend to fall short of a “great” app.&lt;/p&gt;
&lt;p&gt;There are two examples I’m going to talk about. One I’m going to mostly skip over, but it’s huge: threading. Dispatch queues, Swift’s &lt;code&gt;Task&lt;/code&gt;, threads… all very basic concepts in native development. Complicated to use effectively at scale, sure, but the building blocks are quite simple. These are all totally out of the conversation with React Native. (There &lt;em&gt;are&lt;/em&gt; new &lt;a href=&quot;https://docs.swmansion.com/react-native-worklets/&quot;&gt;&lt;code&gt;worklets&lt;/code&gt;&lt;/a&gt; but generally JavaScript is not great at multithreading.)&lt;/p&gt;
&lt;p&gt;So, the other example: fonts. At Wealthsimple, we care a lot about the design and the polish and the user experience of our application, and we had our own font made. Simple Sans. It supports an OpenType font feature called &lt;a href=&quot;https://font-converters.com/guides/opentype-features-guide#case-sensitive-forms&quot;&gt;case-sensitive layouts&lt;/a&gt;. Most fonts render parentheses so they vertically centre with lowercase letters by default. Case-sensitive layouts are a way to raise the parentheses and other punctuation so they align with &lt;em&gt;uppercase&lt;/em&gt; letters instead.&lt;/p&gt;
&lt;p&gt;Default is on the left, case sensitive on the right. See the difference?&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/assets/blog/case-sensitive-forms.pdf&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/case-sensitive-forms-letters.svg&quot; alt=&quot;Case sensitive forms example with letters&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;615&quot; height=&quot;117&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Why would you want that? Most letters are lowercase, so you generally &lt;em&gt;want&lt;/em&gt; parentheses and other punctuation lower to match those letters. But as a financial services app, Wealthsimple renders a lot of numbers. So we made sure our font supported this OpenType feature in order to align the parentheses with the uppercase letters, and therefore with the numbers.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/assets/blog/case-sensitive-forms.pdf&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/case-sensitive-forms-numbers.svg&quot; alt=&quot;Case sensitive forms example with numbers&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;615&quot; height=&quot;117&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;It’s a very small distinction but the parentheses feel too low for numbers otherwise.&lt;/p&gt;
&lt;p&gt;iOS supports this natively, no problem. Android supports it, too. React Native does not. But that’s okay, because I know React Native, and I can just go in and patch it, add boom. Case-sensitive forms in our app, no problem&lt;/p&gt;
&lt;p&gt;It’s kind of an indictment of React Native that I’ve had to add different OpenType features &lt;em&gt;so often&lt;/em&gt; that I could reasonably list this as a skill on my resume.&lt;/p&gt;
&lt;p&gt;This is not an uncommon need: to reach into React Native and extend it, bending it to my will. &lt;em&gt;That&lt;/em&gt; is not “native.” In my opinion, developers should be able to do things like this easily when it’s supported by both iOS and Android.&lt;/p&gt;
&lt;p&gt;But they can’t. Because React Native is not native.&lt;/p&gt;
&lt;p&gt;I eventually started doing React Native at Artsy after I started doing React &lt;em&gt;web&lt;/em&gt; at Artsy, and React is… I think it’s pretty cool, actually! React basically keeps an in-memory store of what your user interface is supposed to look like and then reconciles that to the actual browser DOM on-demand. And if you squint, you can kind of say: okay, React Native has a JavaScript runtime, it’s a virtual DOM, it’s a UIKit view hierarchy.&lt;/p&gt;
&lt;p&gt;React Native is React, right? Well, kind of.&lt;/p&gt;
&lt;p&gt;In one sense, React Native &lt;em&gt;is&lt;/em&gt; React… to a fault. On the web, there are sometimes good reasons to “punch through” the React abstraction to get to the underlying DOM elements, so you can &lt;code&gt;dangerouslySetInnerHTML&lt;/code&gt;, or use a ref, or use an imperative handle, because React &lt;em&gt;has&lt;/em&gt; to work on the web with decades of HTML forms and other quirks. On native, there really aren’t the same escape hatches because React wasn’t built for native.&lt;/p&gt;
&lt;p&gt;There are valid reasons that I might want to “punch through” and do things directly to the &lt;code&gt;UIView&lt;/code&gt; instances. But it’s very difficult because if it doesn’t fit within the web paradigm, then making it easy is not a priority for React. To be fair, that has advantages too: fewer native-specific escape hatches mean that React Native is easier for React web developers to use.&lt;/p&gt;
&lt;p&gt;So React Native &lt;em&gt;is&lt;/em&gt; React. For better and worse.&lt;/p&gt;
&lt;p&gt;I’ve saved the most controversial take for last. The argument I just made was a lie! The JavaScript runtimes are &lt;em&gt;totally different&lt;/em&gt;. In a browser, JavaScript is run by just-in-time compilation to be &lt;em&gt;very&lt;/em&gt; fast. But in React Native apps, we &lt;em&gt;precompile&lt;/em&gt; the JavaScript into bytecode and execute it on a very specific, mobile-optimized JavaScript execution engine called &lt;a href=&quot;https://reactnative.dev/docs/hermes&quot;&gt;Hermes&lt;/a&gt;. Hermes is great, but it’s not a web browser, and you have to know about some of its quirks because it’s so specific.&lt;/p&gt;
&lt;p&gt;React is an active project that comes out with new features, but those features rarely launch on React Native alongside the web. When React &lt;a href=&quot;https://react.dev/reference/react/hooks&quot;&gt;hooks&lt;/a&gt; were released, I remember waiting excitedly for React Native support. Years later, when &lt;a href=&quot;https://react.dev/reference/react/Suspense&quot;&gt;Suspense&lt;/a&gt; launched, I didn’t even bother getting excited because I knew I’d be waiting such a long time.&lt;/p&gt;
&lt;p&gt;So if you’re a React web developer thinking you can just use your React skills with React Native: not exactly. You’re going to bump into differences eventually, and you have to know about these differences to make a great app. I don’t even want to get into the distinction between shadow trees and virtual DOMs.&lt;/p&gt;
&lt;p&gt;React Native is not actually React.&lt;/p&gt;
&lt;h2 id=&quot;the-illusion-of-simplicity&quot;&gt;The Illusion of Simplicity&lt;/h2&gt;
&lt;p&gt;So with all this complexity, you might be surprised to hear that the reason teams sometimes pick React Native is for the sake of simplicity. Unfortunately, that simplicity is an illusion.&lt;/p&gt;
&lt;p&gt;The most compelling reason teams pick React Native (or at least, the one I understand the most) is simplicity. React Native promises to simplify mobile development. But the problem is that there is &lt;em&gt;irreducible complexity&lt;/em&gt; to mobile software.&lt;/p&gt;
&lt;p&gt;Mobile apps get shipped to users’ devices, and updated over time (hopefully). Mobile apps are closer to shrink-wrapped software or on-prem software than modern web apps. Web apps have become the “default” kind of software, so it’s appealing for React Native to flatten the differences between web apps and mobile apps.&lt;/p&gt;
&lt;p&gt;There are many, many differences between web apps and mobile apps. (To name one: application state is not something you have to deal with on the web.) So React Native can’t really simplify that. React Native can only move the complexity around, and as a result, the application becomes much more complex &lt;em&gt;on the whole&lt;/em&gt;.&lt;/p&gt;
&lt;div class=&quot;right-image-container&quot;&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/three-platforms.svg&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/three-platforms.svg&quot; alt=&quot;Swift, Android and Web, showing a complex product organization.&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;720&quot; height=&quot;803&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;To make this concrete: if you’re in North America building software, you likely want an iOS app, an Android app, and probably a web app, probably in React. You’ve got three different teams &lt;em&gt;just on the frontend&lt;/em&gt; trying to build the same product. If you want to release a new feature, you’ve got to keep each of these teams in sync with each other. That’s a lot of overhead, and it slows you down.&lt;/p&gt;
&lt;p&gt;The promise of React Native is: what if we replace all of this complexity with &lt;em&gt;just&lt;/em&gt; React?&lt;/p&gt;
&lt;p&gt;React is easier to hire for. The people making the decision – the directors and the VPs – they often know React already so it’s something they feel familiar with. React Native smooths over the organizational complexity in exchange for adding a ton of technical complexity under the hood.&lt;/p&gt;
&lt;/div&gt;
&lt;p&gt;I have a hard time vilifying this trade-off, but I do think teams sometimes overlook the costs because they’re dispersed throughout the organization and hard to measure.&lt;/p&gt;
&lt;div class=&quot;wide&quot;&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/complexity-charts.svg&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/complexity-charts.svg&quot; alt=&quot;Charts of organizational complexity and code complexity against product complexity, for native and React Native&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;1737&quot; height=&quot;774&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;/div&gt;
&lt;p&gt;As your product gets more complex and you add more features, your organizational complexity with a React Native team grows roughly linearly – assuming you’re architecting your app correctly. But if you have multiple native platforms, you have to coordinate between those teams, and the complexity scales higher.&lt;/p&gt;
&lt;p&gt;From a code perspective, though, the opposite is true: assuming proper architecture, as you scale more products and features on the native side, you can do that without ballooning the complexity of your code. With React Native, you’re forced to. Because if you try to build outside of what it’s optimized for, you’re in for a bad time – and you don’t have to go far outside of the norm to feel the pain. High-performance needs like streaming live data, complex animations, deep platform integrations on iOS or Android… all introduce extreme complexity.&lt;/p&gt;
&lt;p&gt;And even average apps will need to spend time managing the React Native platform. It is not free.&lt;/p&gt;
&lt;blockquote class=&quot;pullquote&quot;&gt;
&lt;p&gt;Businesses tend to “measure what matters.”&lt;/p&gt;
&lt;p&gt;But not everything that matters is easy to measure.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;The prevailing wisdom in business is to measure what matters. You pick something that matters, you measure it, set some goals – OKRs, KPIs, whatever – and use those to make sure what matters moves “up and to the right.”&lt;/p&gt;
&lt;p&gt;But not everything that matters can be easily measured.&lt;/p&gt;
&lt;p&gt;Things like user experience quality are very intangible: they show up across App Store reviews, net promoter scores, or just the feeling users have towards your app. And it’s hard to attribute any of that back to “native versus React Native.” Because these intangible qualities are hard to measure, I think that product teams sometimes underweight the importance of them. And that’s really where React Native suffers.&lt;/p&gt;
&lt;p&gt;But it’s not all bad. Product teams &lt;em&gt;do&lt;/em&gt; like React Native. You get one codebase, one programming language, one-ish deploy process. It &lt;em&gt;is&lt;/em&gt; simpler.&lt;/p&gt;
&lt;p&gt;Testing JavaScript-only changes is really nice too, because you can open up a pull request and have a computer in the cloud generate a JavaScript bundle, post a QR code to the pull request, you scan it with your app, it downloads the bundle, and that’s it. My coworker can test my changes on their phone directly. No provisioning profiles, no signing, nothing. It’s incredible.&lt;/p&gt;
&lt;p&gt;Over-the-air updates are very common with React Native applications, and they’re incredible. I cannot stress enough how amazing it is to fix a production issue with it. I have resolved production incidents by pushing a &lt;a href=&quot;https://github.com/microsoft/react-native-code-push&quot;&gt;CodePush&lt;/a&gt; bundle with fresh JavaScript that gets downloaded to all users, and it works. Everybody does it (&lt;em&gt;or so I hear&lt;/em&gt;) and if you could do it on native mobile apps, I guarantee that you would too.&lt;/p&gt;
&lt;p&gt;Startups and scale-ups need to prioritize speed above all else and that’s okay. They can prioritize speed by using React Native, as long as they know that someday, if the company survives and is successful enough, their needs will likely outgrow React Native. And that’s okay too.&lt;/p&gt;
&lt;p&gt;The real issue I have is that by smoothing out the differences between platforms – between Android and iOS, between web and mobile – it encourages product teams to think about apps as mere software. And apps can be more than just software.&lt;/p&gt;
&lt;h2 id=&quot;the-costs&quot;&gt;The Costs&lt;/h2&gt;
&lt;p&gt;React Native has a huge cost. The dependency upgrade churn is intense. It often comes in lockstep: if you need to upgrade Expo, you’ll need to upgrade React Native, and you’ll also need to upgrade your TypeScript and your Babel and your Reanimated and all of these things. They all have to be done at once, and they’re all constantly moving. Huge risk. It’s a nightmare, and I dislike it more than any other part of building React Native.&lt;/p&gt;
&lt;p&gt;You also get a massively more complex runtime. Every React Native app tolerates a background radiation of random crashes that no one can explain, that are intractable to solve because the stack traces cross the JavaScript-to-C++ boundary, and you have to just live with it.&lt;/p&gt;
&lt;p&gt;It bothers me more and more every day.&lt;/p&gt;
&lt;p&gt;The abstraction of React Native pushes you further away from the platform and the tools. If there’s a user interface idiom or native component that would be really great for your app, well, I hope you’re willing to write some Objective-C++ to reach for it! And if there’s a problem with your application, have fun with the React Native dev tools. The profiling tools have gotten better over time, but it’s hard to hold a candle to Xcode’s Instruments. I actually use Instruments to profile React Native apps because I can get better results from them. But I knew the tools already; teams &lt;em&gt;really&lt;/em&gt; have to push harder to achieve the same quality.&lt;/p&gt;
&lt;p&gt;It’s kind of like the speed of light – the closer you get, the harder you have to push to get even closer. That’s true of native and Swift too, but it gets so much harder, so much earlier, with React Native.&lt;/p&gt;
&lt;p&gt;And unfortunately, React Native has been trying to shrink the size of its core for several years, so some very basic things that you shouldn’t have to use an external library for, you now do. The priority has been to make the core of React Native “more cross-platform” to support Windows, for example. I don’t care about Windows at all. This results in outcomes like React Native not even having an API to access the device’s pasteboard anymore – you have to use a third-party library for that. WTF.&lt;/p&gt;
&lt;h3 id=&quot;personal-costs&quot;&gt;Personal Costs&lt;/h3&gt;
&lt;p&gt;So what about the personal cost to me? It’s complicated.&lt;/p&gt;
&lt;p&gt;Being great at iOS helped me be great at React Native, but I don’t know if that’s because of my technical background or just because I cared. &lt;strong&gt;There is no substitute for caring.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;Unfortunately, React Native pulled me away from iOS: the technical side and the community side. More than I expected. &lt;em&gt;Way&lt;/em&gt; more than I expected.&lt;/p&gt;
&lt;p&gt;But, I’ve gotten chances to solve really interesting technical problems at a scale I probably wouldn’t have been able to without React Native. I also really love TypeScript – it’s a fantastic programming language, and honestly, the way it models types as sets, I really like it.&lt;/p&gt;
&lt;p&gt;However, it’s been difficult to feel like part of the iOS developer community. It’s sometimes been difficult for me to listen to iOS developer podcasts, or read developer blog posts, from peers whom I admire because of how they talk down about React Native and developers like me who use it.&lt;/p&gt;
&lt;p&gt;And the need for me to constantly push React Native harder to reach a great app? It’s left me exhausted. I’m so tired of it.&lt;/p&gt;
&lt;p&gt;So those are the effects React Native has had on me.&lt;/p&gt;
&lt;h3 id=&quot;what-it-does-to-teams-what-it-does-to-apps&quot;&gt;What It Does to Teams, What It Does to Apps&lt;/h3&gt;
&lt;p&gt;I hope I’ve built up enough trust here that I don’t need to describe all the details, but I have compared React Native and native extensively. They have similar-ish ceilings of performance and quality, but React Native has a &lt;em&gt;much&lt;/em&gt; lower floor. What I mean is you can make an app in React Native as good as native if you consistently try hard, but you can also easily make a mistake. Then your React Native app can get a lot worse than it can on native. So that’s what tends to happen.&lt;/p&gt;
&lt;p&gt;Specifically…&lt;/p&gt;
&lt;p&gt;You also just have to accept that React Native is going to chew up &lt;em&gt;at least&lt;/em&gt; 200 megabytes of RAM constantly. That’s the baseline &lt;em&gt;before you even start&lt;/em&gt; rendering anything in your app. And you’ll only stay close to that baseline if you’re profiling often, to keep your app performant. Neither native mobile nor React Native developers tend to profile their apps enough, in my opinion. So, yeah.&lt;/p&gt;
&lt;p&gt;In terms of CPU and graphics performance, you can get &lt;em&gt;close&lt;/em&gt;. I’ve built similarly complex views in SwiftUI and React Native, with streaming data from the network and frequent updates, and tested scrolling performance. The React Native implementation can get within 5 to 10% of SwiftUI. (Even better if you use UIKit, actually, but I’m not here to pick another fight.) 5%, if you squint, is “close enough” to native to be a wash.&lt;/p&gt;
&lt;div class=&quot;wide&quot;&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/ceiling-and-floor.svg&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/ceiling-and-floor.svg&quot; alt=&quot;Native and React Native share the same ceiling, but React Native&#39;s floor is much lower&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;1593&quot; height=&quot;947&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;/div&gt;
&lt;p&gt;So if we visualize this: native and React Native have the same ceiling, but native’s floor is much higher. How bad your app can accidentally be is higher in React Native than in native.&lt;/p&gt;
&lt;p&gt;If you think about an average team reaching an average of what they can accomplish – and I’m not trying to look down on being average, that’s just the way it is – if you look at the spread between the floor and ceilings of native and React Native, it’s &lt;em&gt;possible&lt;/em&gt; to make a great app in React Native, but it’s a matter of organizational priorities.&lt;/p&gt;
&lt;blockquote class=&quot;pullquote&quot;&gt;
&lt;p&gt;Anyone &lt;em&gt;can&lt;/em&gt; make a great app in React Native.&lt;/p&gt;
&lt;p&gt;But they probably won’t.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Is the team using React Native literally &lt;em&gt;always&lt;/em&gt; going to push for the extra effort it takes to make a great app, every time? Teams like that are exceptionally rare.&lt;/p&gt;
&lt;p&gt;The fact is, anyone &lt;em&gt;can&lt;/em&gt; make a great app in React Native, but they probably won’t. And that’s okay. I don’t actually think that every app needs to be exceptional. Some apps can be average. But I’ve only ever wanted to build exceptional software.&lt;/p&gt;
&lt;p&gt;Back when I put up a big fuss about migrating to React Native, Artsy’s Director of Engineering took me aside and promised that we would never compromise the user experience for the sake of using React Native. That we’d always use native if that’s what it took to create a better product.&lt;/p&gt;
&lt;p&gt;And they kept that promise.&lt;/p&gt;
&lt;p&gt;And then they left the company.&lt;/p&gt;
&lt;p&gt;So I was in charge. And so I kept the promise.&lt;/p&gt;
&lt;p&gt;And then I left.&lt;/p&gt;
&lt;p&gt;I have a ton of respect for Artsy. I care about this company and its products a lot. I helped hire and train a lot of the people who are &lt;em&gt;still&lt;/em&gt; there working on this code.&lt;/p&gt;
&lt;p&gt;Are they still prioritizing quality in the same way? I hope so. I don’t know – I can’t know, because I don’t work there anymore.&lt;/p&gt;
&lt;p&gt;This experience really taught me that if you care about the products you’re building, like I care about Artsy, then you have to make decisions with the foreknowledge that you’re not always going to be the one making the decisions. You have to set your replacement up for success, but you also kind of want to box them in a little bit. And I think React Native is a failure to box in a successor.&lt;/p&gt;
&lt;p&gt;So that was Artsy – my time there, my switch to React Native. What about my switch back?&lt;/p&gt;
&lt;h3 id=&quot;steering-towards-native&quot;&gt;Steering Towards Native&lt;/h3&gt;
&lt;div class=&quot;right-image-container&quot;&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/streaming.webp&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/streaming.webp&quot; alt=&quot;Watchlist prices streaming in real time&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;720&quot; height=&quot;1379&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Well, I’m thrilled! &lt;a href=&quot;https://ashfurrow.com/blog/one-year-at-wealthsimple/&quot;&gt;I joined Wealthsimple last year&lt;/a&gt; to bring an unbiased perspective on React Native and native technologies. This is no easy task, to use either stack when it makes sense for the business.&lt;/p&gt;
&lt;p&gt;One of the first projects I worked on was threading. Because JavaScript doesn’t have the capability to put things on a background thread easily, so we built our own. We moved our WebSocket subscriptions into the native layer, with Swift and Kotlin on Android. We have a lot of this type of view in our application – data being updated from WebSockets in real time, and how fast we update the data to the user interface matters to our clients. When we rolled this change out, we saw slow frames (missing the 16-millisecond deadline) drop in half.&lt;/p&gt;
&lt;p&gt;This wasn’t a very complex project. In some senses it was, because we had to be familiar with all these different native programming languages, but on the other hand, it’s just threading. Not “just” threading – there was &lt;a href=&quot;https://swiftrockies.com/recap-2026&quot;&gt;a Swift concurrency talk&lt;/a&gt; the day before mine, and I know concurrency is more complex than I’m describing here.&lt;/p&gt;
&lt;p&gt;But honestly, just moving some data processing to the background thread was a huge step up for our React Native app.&lt;/p&gt;
&lt;/div&gt;
&lt;div class=&quot;right-image-container&quot;&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/context-menu.webp&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/context-menu.webp&quot; alt=&quot;A native context menu on a React Native button&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;1206&quot; height=&quot;1006&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/long-press-menu.webp&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/my-slow-return-to-a-swift-career/long-press-menu.webp&quot; alt=&quot;A long-press context menu on a search result&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;1204&quot; height=&quot;1588&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;I’ve got a couple more examples of ways I’ve integrated native underlying UI components into React Native. A straightforward context menu from a &lt;code&gt;UIButton&lt;/code&gt;, and a long-press &lt;code&gt;UIContextMenuInteraction&lt;/code&gt;.&lt;/p&gt;
&lt;p&gt;The wild thing is that the button is a React Native button, rendered in JavaScript, but it turns into a native context menu. And it all scrolls at 60 or 120Hz, whatever the device supports. It’s performant now, and it supports long press interactions too. (I even worked with an Android developer so it works on both platforms from one JavaScript interface!)&lt;/p&gt;
&lt;p&gt;So we have those nice little touches that make the application feel like it belongs on iOS. Because it does belong on iOS.&lt;/p&gt;
&lt;p&gt;I was hired at Wealthsimple to care about our app like a native iOS developer would, and it’s been a thrill. But it’s been hard, too. I would bet that I’ve struggled more with context menus than any native iOS developer. If you disagree, I want to hear about it.&lt;/p&gt;
&lt;/div&gt;
&lt;p&gt;Personally and professionally, I’m having a blast.&lt;/p&gt;
&lt;p&gt;Unfortunately, I’ve also never been more angry at React Native.&lt;/p&gt;
&lt;p&gt;After I patched our version to support case-sensitive layouts, I did the responsible thing as an open source contributor and opened &lt;a href=&quot;https://github.com/react/react-native/pull/56766&quot;&gt;a pull request to React Native&lt;/a&gt;. I tested it, got an Android phone, found a font with case-sensitive layout on the default system fonts… I did the hard work. And six months later, I haven’t heard a thing.&lt;/p&gt;
&lt;p&gt;Seven years after releasing UIScene in 2019, Apple finally required developers to migrate to it for iOS 27. Apps that link with Xcode 27 without a scene delegate will crash on launch. React Native had &lt;a href=&quot;https://github.com/react-native-community/discussions-and-proposals/pull/967&quot;&gt;an RFC&lt;/a&gt; to do this migration, and when I gave my talk over the summer, there had been no updates all year.&lt;/p&gt;
&lt;p&gt;React Native only released &lt;a href=&quot;https://reactnative.dev/blog/2026/10/19/react-native-0.88&quot;&gt;a version that supported UIScene&lt;/a&gt; in mid-October 2026, a full month &lt;em&gt;after&lt;/em&gt; iOS 27 was released. That means that for that month when iOS 27 was shipping, &lt;em&gt;any&lt;/em&gt; brand new React Native application created would &lt;em&gt;not&lt;/em&gt; be UIScene-based, which means it would not be resizable at all. And React Native still only supports a single scene. That just – it hurts.&lt;/p&gt;
&lt;p&gt;What are the chances React Native apps are going to support the iPhone Duo early after launch? What are the chances React Native apps out there are going to integrate with the new Siri AI features? It’s a joke.&lt;/p&gt;
&lt;p&gt;React Native simply does not care about integrating with its underlying platforms. A friend of mine &lt;a href=&quot;https://github.com/react/react-native/pull/25425&quot;&gt;sent React Native a pull request to add UIScene support in 2019&lt;/a&gt;, and it never got merged in. The project simply does not care.&lt;/p&gt;
&lt;p&gt;I don’t blame any individuals at React Native – these are the same incentives that push companies to operate this way. I understand it. But it’s not for me anymore.&lt;/p&gt;
&lt;h2 id=&quot;wrapping-up&quot;&gt;Wrapping Up&lt;/h2&gt;
&lt;p&gt;We talked about a lot here. I hope you have a greater understanding of how frustrated I am with React Native. Native iOS developers dislike React Native, but &lt;em&gt;I dislike it more&lt;/em&gt;.&lt;/p&gt;
&lt;p&gt;Today we talked about simplicity and the complexity, and how the most complex thing is how I feel about React Native. We talked about how React Native apps will tend towards average, because that’s what React Native is best at. That’s okay, but I’m not okay with that anymore.&lt;/p&gt;
&lt;p&gt;Similar to app quality, I think that my career – in terms of professional satisfaction – has a similar ceiling but a much lower floor. And I’ve felt more satisfied since I decided to be more picky. I’m now only working with teams like Wealthsimple who want my help reaching that ceiling. To create the best app they can.&lt;/p&gt;
&lt;p&gt;As I’ve worked at Wealthsimple I’ve used my skills to create a great app. Often that’s meant building with React Native. But more and more, it means I’m working with Swift.&lt;/p&gt;
&lt;p&gt;That’s it. This is my slow return to a Swift career.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[LLMs Are Tools]]></title>
      <link>https://ashfurrow.com/blog/llms-are-tools/</link>
      <pubDate>Fri, 16 Oct 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/llms-are-tools/</guid>
      <description><![CDATA[I have an enormous respect for Rich Hickey; I think he’s one of the best thinkers among software developers. When I read his post ‘AI’ is Not a Tool (gist), I initially agreed. It felt like a compelling argument. But the more I thought about it, the more...]]></description>
      <content:encoded>&lt;p&gt;I have an enormous respect for Rich Hickey; I think he’s one of the best thinkers among software developers. When I read his post &lt;a href=&quot;https://www.linkedin.com/pulse/ai-tool-rich-hickey-kjl8e/&quot;&gt;&lt;em&gt;‘AI’ is Not a Tool&lt;/em&gt;&lt;/a&gt; (&lt;a href=&quot;https://gist.github.com/svieira/619d60dea942a8d4498cba3bb6b11c02&quot;&gt;gist&lt;/a&gt;), I initially agreed. It felt like a compelling argument. But the more I thought about it, the more I disagreed.&lt;/p&gt;
&lt;p&gt;It still &lt;em&gt;feels&lt;/em&gt; right, but it doesn’t hold up to scrutiny as an argument. Which is really interesting to me, &lt;a href=&quot;https://ashfurrow.com/blog/foresight/&quot;&gt;someone with conflicted feelings&lt;/a&gt; about LLMs but who uses them every day, and who wants to get better at using them.&lt;/p&gt;
&lt;p&gt;This has been eating at me. So I need to get some thoughts out.&lt;/p&gt;
&lt;p&gt;The post tries to define tools, and seems to try (but ultimately fail) to avoid the trap of definition-based arguments: definitions are impossible to agree on. It’s an epistemological distinction, and ultimately not really a valid argument. For any definition of a tool, I could come up with a counter-example: something that is obviously a tool but doesn’t fit the definition, or something that obviously isn’t a tool but &lt;em&gt;does&lt;/em&gt; fit the definition.&lt;/p&gt;
&lt;p&gt;I’m not really going to argue with the post’s definitions of tools, which are vague (good) but somewhat circular (bad). Instead, I’m going to say that &lt;em&gt;I see LLMs being used as tools&lt;/em&gt; every day, and I use them as tools every day. I see them wielded (sometimes surgically and sometimes horrifically). Some developers are good at using LLMs and some are not; I have improved with how I use them to solve problems. Certainly, by some definition, they could be tools?&lt;/p&gt;
&lt;p&gt;Or maybe LLMs are merely “automations” instead, not “real” tools. But here’s the counterexample: if hand saw is a tool, what about a table saw? Table saws are more dangerous and less predictable, and could be similarly described as “automations” rather than tools. But I don’t think anyone would argue they are &lt;em&gt;not&lt;/em&gt; tools in some respect.&lt;/p&gt;
&lt;p&gt;The post tries to make a distinction without a difference, and ends with this somewhat ominous paragraph:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;LLMs are not tools. I know many people use the phrase “just tools” simply to indicate “they are what you make of them”. &lt;strong&gt;Except that’s not true.&lt;/strong&gt; Automation has intrinsic properties and differences from tools that you ignore at your peril.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Based on this post, Rich Hickey clearly doesn’t think highly of LLMs. But the post doesn’t present an actual logical argument. It’s a kind of &lt;em&gt;argument from disgust&lt;/em&gt;. And it was disappointing to read from someone whom I respect a great deal.&lt;/p&gt;
&lt;p&gt;Among software developers with (strong) opinions on AI coding agents, I see two groups emerging:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The first group asks, in horror, “how could you trust code that’s &lt;em&gt;generated by an LLM&lt;/em&gt;??”&lt;/li&gt;
&lt;li&gt;The second group asks, with curiosity, “how &lt;em&gt;could&lt;/em&gt; you trust code that’s generated by an LLM?”&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;I’m in the second camp, and increasingly so.&lt;/p&gt;
&lt;p&gt;(I would have had an easier time getting onboard if AI boosters would &lt;a href=&quot;https://www.businessinsider.com/ai-researchers-smoking-ai-cure-cancer-palantir-cto-2026-10&quot;&gt;stop being so unhinged&lt;/a&gt;. But I digress.)&lt;/p&gt;
&lt;p&gt;As I’ve written about before, &lt;a href=&quot;https://ashfurrow.com/blog/ai-coding-tools/&quot;&gt;I have a hard time relating to software developers who don’t express curiosity towards LLM-based coding agents&lt;/a&gt;. I absolutely relate to having conflicted feelings about them, but I do see an increasing rift here as software developers talk past each other. And I’m not sure what to do about it.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[One Year at Wealthsimple]]></title>
      <link>https://ashfurrow.com/blog/one-year-at-wealthsimple/</link>
      <pubDate>Wed, 14 Oct 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/one-year-at-wealthsimple/</guid>
      <description><![CDATA[One year ago, after reflecting on what I really wanted from my career (and what I didn’t want), I joined Wealthsimple. Time has absolutely flown by! And you know me, I can’t help but blog about it. I joined Wealthsimple with the role of caring about the...]]></description>
      <content:encoded>&lt;p&gt;One year ago, after &lt;a href=&quot;https://ashfurrow.com/blog/work-at-your-spikes/&quot;&gt;reflecting on what I really wanted from my career&lt;/a&gt; (and what I &lt;em&gt;didn’t&lt;/em&gt; want), &lt;a href=&quot;https://ashfurrow.com/blog/joining-wealthsimple/&quot;&gt;I joined Wealthsimple&lt;/a&gt;. Time has absolutely flown by! And you know me, I can’t help but blog about it.&lt;/p&gt;
&lt;p&gt;I joined Wealthsimple with the role of caring about the mobile app as if it were a native iOS app, even though it’s implemented in React Native. After all, it &lt;em&gt;is&lt;/em&gt; a native iOS app, it just happens to use JavaScript to render its UI.&lt;/p&gt;
&lt;p&gt;I cared a lot. And that drove me to accomplish a lot. The first project I helped with cut the stalled frame rate of the app &lt;em&gt;in half&lt;/em&gt;. I felt more accomplished in just the first few months than I had in a few years.&lt;/p&gt;
&lt;p&gt;This summer saw me installing the iOS beta on my main phone and developing new features against the beta SDK. It had been… a while. I felt the joy and pain of a beta OS. Fun! Fun? Fun.&lt;/p&gt;
&lt;p&gt;I navigated a product organization that wasn’t used to concepts like “apps linked against older SDKs will run in compatibility mode on newer iOS versions”, which was challenging. This is the air that iOS developers breathe so it can be tough to work in a a different environment. But I managed to &lt;em&gt;ship so much&lt;/em&gt;. I bridged the gap between the underlying native technology and the JS layer that the organization was familiar with, in a way that empowered my peers. Just &lt;em&gt;doing the thing myself&lt;/em&gt; would have been a failure; instead, I helped others learn and grow. And I learned and grew myself.&lt;/p&gt;
&lt;p&gt;When I joined Wealthsimple, I’d been building with React Native for eight years – fully half my career as an iOS developer – and I was excited to push the bounds of app quality with React Native. I didn’t expect to hit those bounds so easily, so soon. Here’s an example.&lt;/p&gt;
&lt;p&gt;Do you see this context menu? I built it! Looks simple, right? &lt;strong&gt;Wrong!&lt;/strong&gt; This is a native menu invoked from a UIButton, which contains a view sub-hierarchy that gets defined in JavaScript and managed by arcane Objective-C++ wizardry. I’m so proud of it, yet I feel appalled.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/one-year-at-wealthsimple/context-menu.webp&quot; data-lightbox=&quot;&quot;&gt;&lt;img src=&quot;https://ashfurrow.com/blog/one-year-at-wealthsimple/context-menu.webp&quot; alt=&quot;Normal context menu opening animation, invoked by tapping a navigation item at the top of a list, with normal menu items and icons.&quot; loading=&quot;lazy&quot; decoding=&quot;async&quot; width=&quot;600&quot; height=&quot;500&quot; /&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;My thinking a year ago was that &lt;em&gt;organizational priorities&lt;/em&gt; were the main challenge to making a truly great app with React Native, that if a team were just to hire someone like me then it &lt;em&gt;could&lt;/em&gt; be done. To my surprise, when I was given the resources and was empowered to make a React Native app &lt;em&gt;truly&lt;/em&gt; great, I hit some fairly intractable technical hurdles, fairly quickly.&lt;/p&gt;
&lt;p&gt;That’s part of what spurred me to give my conference talk.&lt;/p&gt;
&lt;p&gt;Oh yeah, I gave a talk!&lt;/p&gt;
&lt;p&gt;Years ago, “back when I was Ash Furrow” (as I sometimes joke to my wife, to her chagrin), I gave conference talks. &lt;a href=&quot;https://ashfurrow.com/speaking/&quot;&gt;&lt;em&gt;A lot&lt;/em&gt;&lt;/a&gt; of conference talks. I hadn’t given a talk since covid, since I moved home to Canada. But this summer, I gave a talk at &lt;a href=&quot;https://swiftrockies.com/&quot;&gt;Swift Rockies&lt;/a&gt;. &lt;a href=&quot;https://www.youtube.com/watch?v=UCimuzlrbcc&quot;&gt;&lt;em&gt;My Slow Return to a Swift Career&lt;/em&gt;&lt;/a&gt; was part career retrospective and part technical exploration of what React Native can and can’t do. I shared a lot of what I’d learned at Wealthsimple.&lt;/p&gt;
&lt;div class=&quot;youtube-embed&quot;&gt;
      &lt;iframe src=&quot;https://www.youtube.com/embed/UCimuzlrbcc&quot; frameborder=&quot;0&quot; allowfullscreen=&quot;&quot; loading=&quot;lazy&quot;&gt;&lt;/iframe&gt;
    &lt;/div&gt;
&lt;p&gt;iOS developers… tend not to hold React Native in high esteem, let’s say. When I first showed up at the conference, I felt like I was wandering into a lion’s den (not &lt;em&gt;only&lt;/em&gt; because the conference was held at the Calgary Zoo). But almost immediately, I felt more like I was returning to an old favourite hangout, with familiar and friendly faces.&lt;/p&gt;
&lt;p&gt;“Welcome back”, one of my peers told me after my talk. I fought back tears.&lt;/p&gt;
&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/ai-coding-tools/&quot;&gt;I’d already been using AI coding tools throughout 2025&lt;/a&gt;, but my use of them really accelerated after joining Wealthsimple. I didn’t join specifically because of their approach to AI tools, so this has been an unexpected, welcome highlight of my time here so far. Wealthsimple’s approach to AI code tools has been pragmatic, responsible, and ambitious. My proficiency at using LLMs to ship products is leaps and bounds ahead of where it was a year ago. I’ve learned so much, and I feel closer to the cutting edge of how product development is evolving.&lt;/p&gt;
&lt;p&gt;I’d been worried when I joined that my Swift skills would be too outdated to be a “proper” iOS developer, that I’d somehow tricked Wealthsimple into hiring me. But there’s actually never been a better time to be “rusty” at a technical skill. I wasn’t hired just for what I can &lt;em&gt;do&lt;/em&gt;; I was hired for my opinions, for what I can inspire others to do, and for how I care about my work. &lt;strong&gt;There is no substitute for caring&lt;/strong&gt; and that’s never been as true as it is with the proliferation of AI coding tools.&lt;/p&gt;
&lt;p&gt;So that’s year one! I’ve built a lot and I’ve had a big impact: millions of Canadians benefit from my contributions so far! I’ve made interesting mistakes, the best kind. My sense of professional pride and accomplishment is as high as it’s been since Artsy, “back when I was Ash Furrow.” And I’m thrilled for what I’m working next. Stay tuned!&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[Radiating Intent]]></title>
      <link>https://ashfurrow.com/blog/radiating-intent/</link>
      <pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/radiating-intent/</guid>
      <description><![CDATA[I had a new member of my team join recently and, when I met with them one-on-one for the first time, they complimented me on my standup updates. My team works remotely so standups are an important touchpoint. My colleague and I talked a bit more about...]]></description>
      <content:encoded>&lt;p&gt;I had a new member of my team join recently and, when I met with them one-on-one for the first time, they complimented me on my standup updates. My team works remotely so standups are an important touchpoint. My colleague and I talked a bit more about what made my updates helpful and I realized that I’ve never actually written a public blog post about this.&lt;/p&gt;
&lt;p&gt;This is that post.&lt;/p&gt;
&lt;p&gt;The idea is called &lt;strong&gt;radiating intent&lt;/strong&gt; and it’s guided how I show up and communicating professionally for nearly a decade. (I don’t remember where I heard this from initially.)&lt;/p&gt;
&lt;p&gt;Here’s how it works: I keep the people I work appropriately aware of what I’m about to do. If they have feedback, concerns, or suggestions, I make sure they know that I &lt;em&gt;do&lt;/em&gt; want to hear from them. But I’m not &lt;em&gt;waiting&lt;/em&gt; to hear from them; I’m going to take action by default.&lt;/p&gt;
&lt;p&gt;Radiating my intent has a few key benefits for me and my team:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;It speeds me up because I’m not waiting for approval.&lt;/li&gt;
&lt;li&gt;It makes it easier to work with me, since coworkers only need to reach out if they have something to say.&lt;/li&gt;
&lt;li&gt;It invites my team to give me feedback (if they have any) which helps reinforce healthy team dynamics. Especially when working remotely.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;It’s not a complex idea. “Keep those around you appropriately aware of your next move.” But the word “appropriately” is key: too much detail and I become a distraction, too little detail and I become a lose cannon. And it really is my &lt;em&gt;immediate&lt;/em&gt; next steps, not something I’m planning way in the future.&lt;/p&gt;
&lt;p&gt;I try to think about who I’m talking with and what &lt;em&gt;they&lt;/em&gt; need to know to give me valuable feedback.&lt;/p&gt;
&lt;p&gt;It’s also a cue for me to figure out what &lt;em&gt;I&lt;/em&gt; think I should be doing, rather than relying on others to tell me. Now, you might say “that’s all well and good for a staff developer who gets a lot of autonomy, but it won’t work for junior developers.” And I disagree: I think that having a sense for what you should be doing next at any given time is a valuable skill at any career level. Radiating intent helps you hone that skill. And it shows initiative, which is &lt;em&gt;especially&lt;/em&gt; distinguishing for junior developers.&lt;/p&gt;
&lt;p&gt;Here are some counterexamples and what I would do instead:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;“What ticket should I work on next?” -&amp;gt; “I plan to work on &lt;em&gt;this&lt;/em&gt; ticket next. Let me know if anyone has something else they need.”&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Even at the ticket level, when you find your own next work item on your own it shows initiative. Maybe you get it wrong, but then you get to learn from that. Most managers would rather correct you on this than try to find you a ticket. The extra bit at the end (“let me know if…”) is a hedge in case you made the wrong call. As your experience and responsibility grows, you can use these kinds of hedges less and less.&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;“Do you think we should find help with this project? We’re at risk of missing the deadline.” -&amp;gt; “We’re at risk of missing our deadline so I’ll be asking Andre this afternoon if he has spare time to help.”&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Not only are you making a call here, but you’re being specific with your solution. It’s a bit more senior-level of a call, but this kind of thing happens all the time when building software.&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;“Next week has a holiday, should we get on-call coverage?” -&amp;gt; “Next week has a holiday and I think we need on-call coverage. I’ll let the exec team know that we plan to cover it, and I’ll make sure whoever volunteers knows our lieu-time policy.”&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;People-based decisions like this can be tricky. Maybe this wasn’t the right move! Maybe &lt;em&gt;no one&lt;/em&gt; needs to be on-call during a holiday! You team can tell you. Or, the exec team will since you’ll be radiating intent to them, too.&lt;/p&gt;
&lt;p&gt;Radiating intent works well in small startups and large companies. I know because I’ve used it in both settings.&lt;/p&gt;
&lt;p&gt;I’ve heard the expression that “it’s better to beg forgiveness than ask permission” too often in my career. And I get it. No one wants to get stuck waiting on someone else’s approval, we’d rather move quickly based on our own convictions. But if we keep secrets then we risk misaligning our work with the rest of the company. We risk duplicating work that someone else is already doing, closing ourselves off from feedback that could improve our work, or wasting time on low-priority tasks when something more urgent needs to be done.&lt;/p&gt;
&lt;p&gt;Begging forgiveness sucks. So does asking for permission. I ask for nothing, but I always remain open to suggestions.&lt;/p&gt;
</content:encoded>
    </item>
    <item>
      <title><![CDATA[May You Live Forth and Prosper]]></title>
      <link>https://ashfurrow.com/blog/may-you-live-forth-and-prosper/</link>
      <pubDate>Mon, 04 May 2026 00:00:00 GMT</pubDate>
      <guid>https://ashfurrow.com/blog/may-you-live-forth-and-prosper/</guid>
      <description><![CDATA[Three years ago today, I was laid off from Shopify. Ever since, Star Wars Day has been an annual reminder of that difficult experience. Legally speaking, I have no opinions about my former employer. But… getting laid off was a truly awful experience, and...]]></description>
      <content:encoded>&lt;p&gt;&lt;a href=&quot;https://ashfurrow.com/blog/what-next/&quot;&gt;Three years ago today, I was laid off from Shopify&lt;/a&gt;. Ever since, &lt;a href=&quot;https://www.youtube.com/watch?v=M_ZRsf9sneQ&quot;&gt;Star Wars Day&lt;/a&gt; has been an annual reminder of that difficult experience.&lt;/p&gt;
&lt;p&gt;Legally speaking, I have no opinions about my former employer. But… getting laid off was a truly awful experience, and its timing in my life could barely have been worse. I can’t avoid the company, either: Shopify is a huge part of Canadian tech, I use a lot of Shopify open source software, and Shopify advertises on many of my favourite podcasts.&lt;/p&gt;
&lt;p&gt;So it’s hard for me to have no opinion about Shopify. It’s a daily practice.&lt;/p&gt;
&lt;p&gt;I’m not one of those people who has gratitude for the bad things that have happened to them. There’s still a piece of me that wishes it hadn’t happened. However, I &lt;em&gt;am&lt;/em&gt; grateful for where I am &lt;em&gt;now&lt;/em&gt;. And I wouldn’t be where I am (or who I am) without my experiences. Even the bad ones. Maybe even especially the bad ones. I’ve accepted what happened, and lately, my daily practice of not having opinions about Shopify has felt a little lighter.&lt;/p&gt;
</content:encoded>
    </item>
  </channel>
</rss>