Matt Ström-Awn

The product design talent crisis Why we should invest in the next generation of product designers

New essay! Bad news: the “check engine” light is on for the practice of product design as a whole.

matthewstrom.com/writing/prod...

the prevalence of em dashes in ai writing means my writing is almost certainly in the training data set

Putting it into the universe: any company out there want to hire me to help them build an "associate product designer" program, modeled after many tech co's "associate product manager" programs (apmlist.com)? There's a lot of potential in the market right now.

You know what's better than a 3-legged stool? A 4-legged stool.

(please send biblically-accurate stool memes)

Footnotes - bringing them up to side notes or popovers, but keeping the end-of-page ones for semantics and accessibility

The current desire for “Super IC” designers is possible because of a previous investment in coaches, mentors, and managers, that helped develop those folks.

As companies cull managers and hire “player-coaches”, the next generation of ICs will lose out on much-needed development opportunities.

Is there anyone out there doing AI for design using computer vision (segmentation, world model, etc) instead of using language (LLMs, transformers, etc)?

definitely, and also depends on the relative productivity of what you're switching between.

It's true that the coding takes longer, but there's some extra productivity due to what you can do while the agent is working.

Based on a few weeks of using a coding agent, i think the recent result that coding AI slows developers comes down to two factors:

1. Coding agents allow the developer to multi-task
2. Multi-tasking involves context switching, which has a penalty

troubleshouted

If you're working with design tokens, you might have noticed that the color format has been updated in the latest editor's draft at tr.designtokens.org/color/. But updating your tokens files is a snap! Here's a node script you can run that will get you up to speed: gist.github.com/ilikescience...

They’re calling him the Fizzler

I was an early adopter! Looks like I bought a license in 2013. I think it was the first time I used markdown.

Keep doing great stuff, and lemme know if you ever need design help :)

And if you don't need any of those things, but wanna work together anyway, you should still get in touch :)

After ~10 years of in-house design leadership, I'm going back to my roots as an independent designer. If you need:

- 0 → 1 design partnership
- Mentorship, coaching, or design org design
- Early stage design strategy, especially in AI and Fintech

Get in touch: hello@mu.design

Worst,

Kirby is a gem, and @bastianallgeier.com has been at it consistently for … more than a decade? I love a small team building a great product.

I can trace my entire career back to my ability to customize the scrollbar of my friends' xanga pages

this is my favorite genre of explainer essay: "you can just do things"

soupçon hits the triple word score of inkhorn words: it's borrowed from french, it has a funny accent, and it's easy to mispronounce (bonus word nerd shibboleth)

also good joke

Regardless of the aesthetics of liquid glass, I think the enduring change in ios26 is that chrome (tab bars, url bar, etc) no longer has a background, and content extends edge-to-edge — @itsdpark.bsky.social‬ was ahead of the curve with their 'no tab bars' manifesto: itsdpark.com/research/tab...

Shdr: a lil WebGL library

It’s a simple, slightly-opinionated WebGL library called Shdr. It won’t help you write shaders, per se, but it will help you render them without a lot of boilerplate or ceremony. I don’t know if it will be useful to anyone else, but it was useful to me.

Check it out here: shdr.andystew.art

(yes, gpt-image-1 generated an image with a circular crop 'baked in' 🤦‍♂️)

A grid of images generated by AI. The title reads "Prompt: 'A user's profile picture'". The rows are labeled 'gpt-image-1', 'Google Imagen 3', and 'Titan v2'. The images are stylistically consistent in each row.

Doing some experiments with Figma's AI image generation; surprising results! Images generated by each model vary little within a consistent prompt. While the current image generation features might currently fit into graphic-design-type use cases, I don't see how they'd fit into UI design.

I often see design craft positioned as a reaction to the ubiquity of design systems. As in: great designers know when to break out of the system in search of higher quality. But there's a convergence that happens at high levels of systems thinking; exceptional systems result in well-crafted UI.

i've always taken it as a warning

Looking for any design help? Doing UI systems for Terminal would be a dreamy gig.

1. Depending on where you do business, it may be required by law
2. No matter where you do business, it is the right thing to do

Just polishing up my next LinkedIn post, “Design is a Dharmic wheel in a constant state of death and rebirth”