A few weeks ago I published a deep dive into the multi-faceted design problems AI is exposing. There are points of leverage up and down the stack, each affecting the end user experience and contributing value. It’s an incredible time to be a designer.
Sitting in a coffee shop that is playing late 90s radio pop. Two fellow millennial women at separate tables both started singing the lyrics under their breath, while two K-12 girls at the register sang them outloud to each other. Backstreet's back, apparently (here for it)
just spent 20 mins getting mad at the robot in the box only to realize I had inadvertently kicked off a tougher task while still in Luna. (1) the UX of model routing sucks (2) I feel like a moron (3) sorry I yelled at you bro but c'mon your parents should have taught you better
It’s Nov 4 2008, Rihanna’s Live Your Life is at the top of the charts, Single Ladies is about to drop, you have tickets to see Twilight on opening night, we’re saying goodbye to Bush jr and endless war, and feeling hopeful
Drag-and-drop makes the digital graspable, but building it well is fiddly enough that many don’t bother. Declarative Dragging (with Josh Horowitz & Jeff Heer, #uist2026 best paper award), lets you make an element draggable just by listing the states dragging it can reach.
This seems immensely short sighted by Google. I understand making tradeoffs to follow a market but using its dominance to coerce AI adoption in K-12 is a big self own. Plus the likely consequences are MORE people seeing AI as “bad,” doing an overall disservice to their industry
Other species play, mate, use tools, and create beautiful things on instinct, but I can't think of another species on earth that deliberately crafts or alters its surroundings for the purpose of making life easier and more enjoyable for strangers. Design is distinctly human.
Design in the wilderness
There was a time 10 or so years ago when we got really focused on human centered design. I think it's time to retrieve that earnestness. When we were designing web sites we could perfectly architect every touchpoint.
Then apps, with their algorithms, where we did our best to provide users a map but they ultimately chose their own journeys.
Now, they are in the wilderness, and we can't possibly craft every potential experience or interaction they will encounter. So we have to think differently.
I'm increasingly seeing experience design merge with service design. In the former, we shaped the thing itself. In the latter, we shape the inputs, improve the connections, make it easy to pass data out and make it legible and true when it's received, reduce inefficiencies, improve predictability, always focused on the humans in the center.
I actually don't see agents as our new users. They are intermediary software, like RNA. It's important they are successful in their task, but I'm not willing to sacrifice the importance that the ultimate persona in the matrix is the human.
So in this context, whether you're working on interfaces, or agentic workflows, or model behavior, we're all working in service of the people trying to derive value, efficiency, and meaning out of the things we are creating.
Because honestly, technology sometimes feels like a pretty meaningless industry to work in. Staying centered on what people need, and how I can make their lives easier and better is what drives my engagement and my ambition
"I actually don't see agents as our new users. They are intermediary software, like RNA. It's important they are successful in their task, but I'm not willing to sacrifice the importance that the ultimate persona in the matrix is the human.
So in this context, whether you're working on interfaces, or agentic workflows, or model behavior, we're all working in service of the people trying to derive value, efficiency, and meaning out of the things we are creating."
Design in the wilderness
There was a time 10 or so years ago when we got really focused on human centered design. I think it's time to retrieve that earnestness. When we were designing web sites we could perfectly architect every touchpoint.
Then apps, with their algorithms, where we did our best to provide users a map but they ultimately chose their own journeys.
Now, they are in the wilderness, and we can't possibly craft every potential experience or interaction they will encounter. So we have to think differently.
I'm increasingly seeing experience design merge with service design. In the former, we shaped the thing itself. In the latter, we shape the inputs, improve the connections, make it easy to pass data out and make it legible and true when it's received, reduce inefficiencies, improve predictability, always focused on the humans in the center.
I actually don't see agents as our new users. They are intermediary software, like RNA. It's important they are successful in their task, but I'm not willing to sacrifice the importance that the ultimate persona in the matrix is the human.
So in this context, whether you're working on interfaces, or agentic workflows, or model behavior, we're all working in service of the people trying to derive value, efficiency, and meaning out of the things we are creating.
Because honestly, technology sometimes feels like a pretty meaningless industry to work in. Staying centered on what people need, and how I can make their lives easier and better is what drives my engagement and my ambition
Ok let's run this old poll back again and see if things have changed…
Is it easier to teach/learn visual design, or product strategy?
Like if someone had a gap in one… which is typically more coachable?
Design in the wilderness
There was a time 10 or so years ago when we got really focused on human centered design. I think it's time to retrieve that earnestness. When we were designing web sites we could perfectly architect every touchpoint.
Then apps, with their algorithms, where we did our best to provide users a map but they ultimately chose their own journeys.
Now, they are in the wilderness, and we can't possibly craft every potential experience or interaction they will encounter. So we have to think differently.
I'm increasingly seeing experience design merge with service design. In the former, we shaped the thing itself. In the latter, we shape the inputs, improve the connections, make it easy to pass data out and make it legible and true when it's received, reduce inefficiencies, improve predictability, always focused on the humans in the center.
I actually don't see agents as our new users. They are intermediary software, like RNA. It's important they are successful in their task, but I'm not willing to sacrifice the importance that the ultimate persona in the matrix is the human.
So in this context, whether you're working on interfaces, or agentic workflows, or model behavior, we're all working in service of the people trying to derive value, efficiency, and meaning out of the things we are creating.
Because honestly, technology sometimes feels like a pretty meaningless industry to work in. Staying centered on what people need, and how I can make their lives easier and better is what drives my engagement and my ambition
I'm continually fascinated by the different ways models approach problems and express themselves. They are truly distinct in ways we are only just starting to see.
Like, check out this research from @lu__jasperjasperlu.com/blog/rl-design-…, which explores the downstream effects of having Sol vs. Sonnet train a visual design model. Each model ended up training distinct behavior into its subjects off of the same training set (though notably, both models when set as judges preferred Sonnet's lineage)
I continue to be skeptical that we can train inherent taste into the model's instincts, at least as far as the current generation is concerned. Taste is too contextual to be one-shotted unless drastically overfit to a very specific dimension.
But this does raise some interesting questions around how we might blend models in a harness to evoke tasteful behavior. If one model family stands out in its ability to train on diversity of creative directions, while another is most capable in applying fundamentals, you could see how we could get closer to having a model that can both express novelty, and manage visual quality, at least over the course of a generative workflow vs. a 1-shot solution.
personally, I see Taste as differentiated from trend-oriented judgement or knowledge of fundamentals alone:
- I define Taste as the ability to make distinctly ”Good” choices relative to the context without relying on the influence or instruction of others.
- “Good” is some function of understanding fundamentals and deciphering trends. It takes something extra to connect this “good” intentionally to the graph of context
- To be a taste maker is even more distinct than just the ability to be tasteful
And to Jasper's point, there's a third dimension of Taste that is personal, such that there can be legitimate disagreements between two experts or taste makers within the same scenario.
In all, there's something deeply familiar about building a team of T-shaped members, with strong creative direction at the front. I don't care that a model is tasteful. I care that it can product objectively good results while following my taste, so I can focus on steering it.
Come for the meta analysis of modern practices, stay for the rest diving into the quality of a distribution-normalized concept of a chair.
“What we have here is, quite literally, a chair designed by someone who has read a lot about chairs but never actually sat in one.”