https://www.justinwetch.com/blog/ AI tools, interface design, product taste, and project writeups by Justin Wetch. en-us Tue, 02 Jun 2026 16:43:15 +0000 https://www.justinwetch.com/blog/skillrsi/ https://www.justinwetch.com/blog/skillrsi/ Tue, 02 Jun 2026 16:43:15 +0000 Recursive self-improvement is one of the big ideas in AI right now. You can see the shape of it in projects like Andrej Karpathy’s autoresearch, where an agent changes code, runs a short training experiment, checks whether the metric improved, and repeats. The reason that loop can run is that it has a number to optimize. Training loss goes down or it doesn’t. https://www.justinwetch.com/blog/swipeship/ https://www.justinwetch.com/blog/swipeship/ Wed, 20 May 2026 20:52:43 +0000 SwipeShip is a tiny iPhone remote for vibecoding. When your AI coding agent hits a taste question, a card appears on your phone. You swipe yes or no, optionally drop a comment, and the agent keeps building. The whole interaction is as easy as swiping on Tinder. https://www.justinwetch.com/blog/higagents/ https://www.justinwetch.com/blog/higagents/ Tue, 17 Mar 2026 04:24:14 +0000 Ask your coding agent to follow Apple's Human Interface Guidelines and watch what happens. It'll give you something that sounds right. Confident language, plausible-sounding specs, reasonable defaults. But the measurements will be wrong. The platform distinctions will be invented. The specific rules, the ones that actually matter when you're shipping an app, will be hallucinated with the same fluency as everything else. https://www.justinwetch.com/blog/handtype/ https://www.justinwetch.com/blog/handtype/ Sat, 07 Mar 2026 20:40:36 +0000 HandType is a gesture-controlled typography workstation. You point a webcam at your hand, rotate your palm, and font parameters change in real time. Open palm rotation adjusts the active modifier. Peace sign cycles through the alphabet. Five modifiers (weight, slant, width, pixel, corner radius) let you sculpt any letterform procedurally, and you can upload any TrueType, OpenType, or WOFF file to work with. Everything runs client-side in the browser. https://www.justinwetch.com/blog/psychohistory/ https://www.justinwetch.com/blog/psychohistory/ Sun, 01 Mar 2026 17:17:00 +0000 Growing up in Alaska, my bus ride to school was about an hour each way, longer in winter. I didn’t have a cell phone, so I read a lot of sci fi books. The books I kept coming back to were Isaac Asimov's Foundation series, not just Foundation itself but the extended universe, which remains some of my favorite science fiction. I'd read on the bus to school and on the bus home, and somewhere in those hours the idea of psychohistory lodged itself permanently in my brain: that the future of human... https://www.justinwetch.com/blog/mtom/ https://www.justinwetch.com/blog/mtom/ Mon, 09 Feb 2026 06:35:52 +0000 Last month I was reviewing Anthropic's open-source frontend design skill , a system prompt meant to help Claude generate distinctive web interfaces, and I found an instruction that stopped me cold: "NEVER converge on common choices across generations." https://www.justinwetch.com/blog/colorsculpt/ https://www.justinwetch.com/blog/colorsculpt/ Fri, 06 Feb 2026 19:55:02 +0000 Around 2007, someone on the Houdini forums posted a 3D LUT viewer. A LUT is a look-up table, essentially a mathematical recipe that transforms colors in an image (the thing that gives a movie its "look," that teal-and-orange blockbuster palette or that faded film warmth). This viewer took that recipe and plotted it as a cloud of points in three-dimensional space. Every color became a coordinate. Every color shift became visible as physical movement. The entire logic of a color grade, normally... https://www.justinwetch.com/blog/naturalphoto/ https://www.justinwetch.com/blog/naturalphoto/ Tue, 03 Feb 2026 22:50:18 +0000 Growing up, my mom was a professional photographer. She taught me everything she knew getting started in Alaska, where I spent my childhood shooting landscapes in wilderness that punished bad exposure decisions. Later I got into cinematography, then 3D rendering, and somewhere along the way I became the kind of person who has strong opinions about color science, who can tell you the difference between Portra 400's highlight roll-off and Kodak Vision3 500T's shadow response, who built a personal... https://www.justinwetch.com/blog/skilleval/ https://www.justinwetch.com/blog/skilleval/ Tue, 13 Jan 2026 04:37:01 +0000 Last week, I wrote about improving Anthropic's frontend design skill . The short version: I noticed the skill file had instructions Claude couldn't actually follow, rewrote it with clearer guidance, and needed a way to prove the changes were better beyond my own personal taste. So I built an evaluation system, ran 30 blind comparison pairs across the 3 levels of the Claude 4.5 model family, and found the revised skill won 75% of decisive matchups. https://www.justinwetch.com/blog/claudecodeide/ https://www.justinwetch.com/blog/claudecodeide/ Sat, 10 Jan 2026 19:56:53 +0000 Claude Code has become one of the most powerful tools in my workflow. It writes code, debugs problems, navigates complex codebases, and does things that would have seemed like science fiction two years ago. But it lives in a terminal. And I'm not a terminal person. https://www.justinwetch.com/blog/improvingclaudefrontend/ https://www.justinwetch.com/blog/improvingclaudefrontend/ Mon, 05 Jan 2026 20:46:46 +0000 After building a GUI for Anthropic's Bloom framework last month (and more recently, proposing ten new behaviors it could test for), I found myself wanting to dig deeper into their ecosystem. The Bloom project gave me a lot of respect for how Anthropic approaches tooling and evaluation, and I was curious what else they had open-sourced. https://www.justinwetch.com/blog/10behaviorsbloom/ https://www.justinwetch.com/blog/10behaviorsbloom/ Fri, 02 Jan 2026 07:19:00 +0000 After recently building a GUI for Anthropic's Bloom and thinking through the implications of ASI on democracy , I've been looking for ways to contribute more directly to alignment work. Bloom is Anthropic's open-source framework for evaluating model behaviors, a systematic way to test whether models exhibit patterns like deception, sycophancy, or resistance to correction. Building the interface got me familiar with the framework's structure, and that familiarity made me want to think about what... https://www.justinwetch.com/blog/spatialskeuomorphism/ https://www.justinwetch.com/blog/spatialskeuomorphism/ Sat, 27 Dec 2025 00:53:35 +0000 The word "interface" gets used so often we forget it’s a literal thing: a boundary. An interface is a boundary where two systems meet, and in software, those systems are: arbitrary digital logic on one side, and a biological processor with 300 million years of spatial heuristics on the other. https://www.justinwetch.com/blog/anthropicbloom/ https://www.justinwetch.com/blog/anthropicbloom/ Mon, 22 Dec 2025 03:52:11 +0000 Anthropic open-sourced Bloom two days ago, a framework for testing LLMs for concerning behaviors like sycophancy, political bias, and self-preservation. The underlying system is elegant: a four-stage pipeline that generates test scenarios, runs conversations with target models, and scores behavior presence with cited evidence. It's a metacognitive scaffolding approach that works beautifully for this kind of evaluation (for more on why I think scaffolding is so powerful, see my piece on... https://www.justinwetch.com/blog/interfacedesign/ https://www.justinwetch.com/blog/interfacedesign/ Sun, 21 Dec 2025 19:18:21 +0000 Dieter Rams gave us ten principles for good design. The Gestalt psychologists mapped how the brain organizes perception. What follows is my attempt to synthesize these traditions and my own experience and opinions into something I can continually reuse and refer back to. Not a checklist, but a way of seeing. Ten ideas that explain why some interfaces feel right and others feel like work. https://www.justinwetch.com/blog/democracyasi/ https://www.justinwetch.com/blog/democracyasi/ Thu, 18 Dec 2025 06:52:43 +0000 This is a thought experiment, not a prediction or a policy proposal, but an attempt to sketch one possible good outcome so we have something to argue against and refine. I'm probably wrong about most of this, but I think being wrong specifically is more useful than being vague. https://www.justinwetch.com/blog/scaffold/ https://www.justinwetch.com/blog/scaffold/ Sun, 14 Dec 2025 08:19:22 +0000 GPT-4 was clearly intelligent, you could feel it. But ask it to write a book and it would fall apart by chapter two. Not because it lacked the ability to write well (with the right prompting), but because it couldn't remember what it was doing. The context window would fill, the model would lose the thread, and you'd end up with something that read like it was written by a brilliant amnesiac. https://www.justinwetch.com/blog/wdoaessay/ https://www.justinwetch.com/blog/wdoaessay/ Mon, 01 Dec 2025 05:35:57 +0000 Most artist profiles read like resumes. Born here, studied there, exhibited at these places. It's a timeline of achievements, but it rarely makes you feel anything. The information is there, but the story isn't.