Been getting asked a lot lately: what do you think about the current design shift?
That’s kind of a million-dollar question. But I also think it’s a critical product thinking question. Everyone will answer it differently depending on what they’re building, who they’re building for, and how deep they are in AI products right now. I just wanted to write down my own thoughts, especially on where products seem to be heading, and what that means for design.
A lot of AI products right now feel like they’re showing up in different shapes — agents, workflows, knowledge or skills systems. I’ve been trying to think about the design shift across those, and for me it sits in layers. Higher level stuff shapes the lower level stuff.
The product-level shift
For a long time, a lot of product design was about making things easier for humans to use like clearer flows, less friction, better defaults, faster paths to an outcome.
That still matters. But with AI products, I feel the center is moving. It’s less just “help a person get through a flow,” and more “help a person work with AI so they can do more, and still stay in control of what comes out.”
So the job becomes something closer to: communication + amplification + control.
Not only designing the steps a user takes, but designing how they talk to the system, how the system extends what they can do, and how they can still steer, check, and own the outcome.
When I look at products through that lens, two things feel especially critical.
First: how well the product actually integrates AI into the work. Not “AI exists somewhere in the UI,” but whether it helps people get more done in a real way with less time, less wasted credits, less back-and-forth that doesn’t go anywhere. Efficiency here is not just speed. It’s whether the collaboration with AI is actually useful.
Second: the cost of trust. Can people trust the source of truth? Is the context accurate enough? Does the system know what it should know, and not pretend about what it doesn’t? And where are the boundaries of authorization like what it can access, what it can change, what it should never touch without a human?
If those two are weak, the product can look smart and still feel expensive, risky, or tiring to use.
So yeah. I think the product-level shift is not that ease of use stopped mattering. It’s that the job got bigger: still easy to use, and also easy to work with AI, making more outcome, and stay in control of that outcome.
The paradigm-level shift
Under that, the way people get work done inside these products is also changing.
It often starts with a first prompt. Sometimes with suggestions or templates to help starting. Then maybe a plan, sometimes optional, sometimes really necessary depending on how high-stakes the task is. Then resources and connectors — tools, files, data, other systems the agent can actually use. Then maybe a preview or draft, which feels more and more important because every run has a cost, and people want to feel the direction before burning too much.
Then comes the part I care about a lot: outcome control. Review. Iteration. Options and variations. Being able to go again without starting from zero. Being able to edit, interrupt, redirect basically let the human interfere when needed.
And then, if this is meant for real work, it has to get closer to something shippable. Format, channel, adaptability. Not just a nice answer sitting in a chat. Something that can leave the product and live in the world.
Then there’s the long-term layer that a lot of demos skip: automation, maintenance, observability, tracking, revamping. Because if something is useful once, cool. If people are going to rely on it, it has to keep working as context changes.
So the paradigm underneath is less “one screen, one task” and more “a loop you can enter, steer, and improve.”
I don’t think every product needs every step. But I do think this loop is becoming the shape of a lot of AI product experiences, whether teams name it that way or not.
The interaction and visual shift
And then of course the surface is shifting too.
We’re seeing all kinds of collaboration interfaces right now: node-based and procedural stuff, canvases, in-browser editing, CLI, plain input fields, voice, shortcuts etc. Probably more by next week. It feels like everyone is still exploring what “working with AI” should look like for different people and different scenarios.
I don’t think there’s one winning interaction model yet. A canvas might be perfect for one kind of thinking and completely wrong for another. A chat box might feel simple, or it might hide too much control. A node graph might feel powerful, or just heavy. Same for voice, same for CLI.
This layer is changing the fastest, and honestly it’s also the easiest layer to get distracted by. New products every day. New patterns every day. So I’m trying not to over-conclude here. I might write about specific product types later, once I’ve spent more time inside them.
For now, my point is mostly: interaction and visual design still matter a lot, but they should be downstream of the product and paradigm questions. Otherwise we just keep inventing prettier ways to do the wrong loop.
What still doesn’t change
Higher level affects lower level. That part feels true to me.
But underneath all of this, I still think the same thing designers have always had to do: understand the user’s context and needs, then design and iterate from there.
What’s different is the speed, and how unfinished everything still feels. There is so much ambiguity. So many possible shapes. So much to test. And the real bar is not whether something demos well, it’s whether it can actually land inside someone’s workflow and stay there.
The surface keeps changing. The loop keeps changing. But the work is still: put AI in the right place in a real person’s workflow, and give them enough control that they can trust the outcome.
Everything else still feels like it’s being figured out in public. I just wanted to put my current point of view somewhere, and hopefully talk to more people about it in this miracle time right now! If you have a different read on the shift, I want to hear it!