Built a live design panel into Icon Forge, a little icon rendering app I’ve been working on. I can now drag sliders to tune radius, shadows, and size, then copy the values as JSON and paste them back into Claude to bake into the code.
Also went deep on Liquid Glass, which is now threaded through most of the interface.
Last thing for tonight.
I’ve been building a little Mac app for design critique.
You drop screens onto a canvas, it generates a full critique, then breaks the feedback into cards pinned to the exact UI elements they reference.
Every finding cites a 600 entry UX knowledge base, and you can watch the camera move screen to screen while it works.
Still janky in spots, pinning is hit and miss, but it’s a start
Quick look at an early SwiftUI project I’ve been exploring with Claude Fable. Still rough, but fun to see how quickly you can create something with this new model.
This is 1 of 3 SwiftUI apps I built tonight with Claude Fable.
Describe a character, get a crisp 1-bit pixel portrait: Floyd-Steinberg dithering, lockable traits, editable prompts. Went from empty folder to working app in ~90 minutes.
The more AI can generate anything, the more taste matters.
Not taste as aesthetic preference. Taste as the discipline of saying no.
AI doesn't have taste. It has pattern. It will give you the default, the thing that offends no one and means nothing. It will give you 100 versions of fine.
Most design already fails not because someone made the wrong choice. It fails because no one made a choice at all. AI makes that failure easier and faster.
Finished Crimson Desert. The story is okay, the world is unforgettable. I lost track of hours just exploring Pywel. 228 hours later, finished the story. Best open world I've gotten lost in since Red Dead.
This is a oversimplification: Pre AI we assumes UI is the unit of value.
Problem → solution → screen. Screen was the deliverable.
Feels like the world is shifting where the unit of value might be the agent's behavior, and the screen is a side effect, if needed at all.
Design shrinks when you start filtering ideas against what's currently allowed.
Drew a new pattern at work. The team said partners won't like it. Kept pushing on it. Months later it's showing up in prototypes. The job isn't just what you ship. It's what you make possible.
@joshpuckett What excites me most, we're entering a point in time where we can build tools that fit our vision, instead of reshaping our vision to fit the tools we have.
been messing around with node-based UIs. Decide to built a sprite generator to see if I could. describe a character, pick some actions, hit run, you get an animated sprite sheet.
been thinking about how much health data sits locked in Apple's ecosystem. built a little iOS app last night that syncs my HealthKit metrics to my home server in the background. resting HR, HRV, sleep, blood pressure. Now I can use the data however I like.
@kwonlabs started with keyword rules, didn't like constantly maintaining them. now an agent watches what i actually read, learns from that instead. It's more proactive with suggestions and aligns to what I'm actually interested in. Well that is the hope. Still testing.
Why pay for Feedly when you can build your own? Built Signal, an AI native RSS reader that generates a daily digest of what actually matters, lets you search everything you've read by meaning (not keyword), and discovers new feeds based on your reading patterns.