The Point of Design Was Never the Pixels

AI makes it dramatically cheaper to put an idea into the world. For designers, that might be one of the best things to happen to the profession.

Why am I writing this?

There’s a popular argument floating around design right now that goes something like this:

AI is making it easier for everyone to design, prototype, and write code, so designers need to move upstream. Stop worrying so much about making things. Become more strategic.

I agree with the second part. I’m not sure I agree with the first.

Designers should absolutely be making things. In fact, I think we should be making MORE things, because the point of the artifact was never really the artifact. It was the reaction.

Making is getting ridiculously fast

I worked on Docusign’s Agent Studio from the beginning with AI-assisted prototyping tools in the loop, and that mattered. Instead of debating an interaction for three days, I could get several ideas onto the board, make them real enough to click through, and give the team something concrete to react to.

Clicking through something is fundamentally different from staring at a static screen. You can feel it. Does this interaction behave the way I expected? Did I know where to click next? Does the transition make sense? Does the whole thing, for lack of a more scientific term, vibe?

Figma’s 2026 research suggests this way of working is spreading quickly. The share of designers participating in development jumped from 21% to 41% in a year, while developers doing design work increased from 44% to 60%.

Figma looks at that and argues that when AI can make everyone dramatically more productive, personal productivity stops being the thing to optimize for. I’d frame it a little differently.

Speed still matters enormously. The mistake is assuming that what you’re accelerating is pixels.

Artifacts are decision-making infrastructure

I frequently see an engineer struggle to understand a PRD until someone draws the thing. I see product teams talk past each other about a roadmap until someone turns it into a model. I see executives disagree abstractly for a week and resolve the disagreement ten minutes after looking at a prototype.

People don’t always have the motivation—or frankly the attention—to construct the same mental model from five pages of text. A designer can give everyone something to point at. That makes the artifact incredibly valuable even when the artifact is wrong. Sometimes especially when it’s wrong. A bad prototype can produce a very useful response: No. Definitely not that.

Great. We just made a decision.

AI lets me get to that decision faster. On Agent Studio, we could build an interaction, put it in front of people, and quickly determine whether our assumptions held up. With modern unmoderated research, a designer can conceivably go from an idea in the morning to a functioning prototype to several pieces of user evidence remarkably quickly.

That changes the economics of conviction. Instead of saying, “I think this is the right interaction,” I can increasingly say, “I think this is the right interaction. Here are the alternatives we explored, here’s what happened when people tried them, and here’s why I’m recommending this direction.”

My grandma is part of my design system

One of my unofficial heuristics is: Could my grandma figure this out?

My grandma isn’t an enterprise software expert. If something requires a detailed understanding of how the system was engineered before the interface makes sense, we may have a problem. By coincidence, while we were working on Agent Studio, we had a bring-your-child-to-work day and one of our product leaders’ kids tried some of our prototypes. That created an unexpectedly useful counterpoint.

On one end of my imaginary usability spectrum was my grandma: someone who isn’t particularly interested in learning the mental model of complex software. On the other was a kid from Gen Alpha: completely comfortable poking around an iPad, but without years of exposure to the strange conventions we’ve collectively decided are normal in enterprise software.

We weren’t pretending this was rigorous user research. It was something simpler: an intuition stress test. What did they try to click? What did they expect to happen? Where did they hesitate?

Modern prototyping makes these tiny experiments almost free, and I think designers should take advantage of that.

Building faster makes taste more important

There’s an uncomfortable corollary to all of this: if a product manager can open an AI tool and generate a more compelling interface than you can, that’s a problem. Not because the PM used your tool, but because your tool was never supposed to be the moat.

Taste was.

Adobe found that explicit AI requirements in U.S. creative job postings increased from 10% to 15% between September 2025 and April 2026. Among mid-market and enterprise employers, 18% of postings explicitly requested AI skills, compared with 8% among smaller businesses.

I don’t interpret that as evidence that designers are becoming less valuable. I interpret it as evidence that proficiency with these tools is becoming less differentiating. Eventually, saying “I know how to use AI” may sound a little like saying “I know how to use Slack.”

Good. I would hope so. The interesting question becomes what you can do with it.

The strongest designers think one layer ahead

This is where I think the gap between good and exceptional designers becomes more visible. Say the assignment is to design a way for someone to create an AI agent. It’s very easy to spend all of your energy perfecting that creation experience, and we should care about that. Make the interactions buttery. Clarify the hierarchy. Fix the empty states. Sweat the language. Make creating the thing feel effortless.

But while I’m designing that, another part of my brain is already somewhere else. What happens when a company has sixty of these? Who owns them? Who can edit them? How does an administrator understand what they’re doing? How do we audit an agent’s decisions? How do we evaluate accuracy, and what happens if that accuracy drifts six months from now? How does governance and permissions work? What happens when two agents overlap?

The interaction we’re designing today is being evaluated against the system it could become tomorrow.

That’s systems thinking. And it matters because product organizations understandably spend enormous amounts of energy trying to land the airplane currently approaching the runway. Engineering is trying to ship it. Product is managing scope. Marketing is figuring out how to explain it. Leadership wants to know when it launches.

The designer has an opportunity to look past the runway and ask where we’re flying next.

AI gives me more room to do that because I can spend less time laboriously constructing the first prototype. Then I can build the next layer too.

Sometimes the prototype should become the spec

There’s another change I don’t think we talk about enough: AI-generated prototypes can increasingly communicate more than appearance.

A designer can explore the interaction, run accessibility checks, specify tab order, identify ARIA requirements, document component behavior, and output enough structured information that Product and Engineering can start reverse-engineering actual requirements from the prototype.

The artifact stops being “Here’s roughly what this should look like” and starts becoming “Here’s how I think this system should behave.”

That’s a meaningful expansion of design’s influence.

There’s broader evidence that judgment—not simply output—is becoming the constraint. Microsoft’s 2026 Work Trend Index found that when AI users were asked which human abilities become more important as AI takes on more work, quality control ranked first at 50%, followed by critical thinking and judgment at 46%. Eighty-six percent said they considered AI’s output a starting point and remained responsible for the thinking.

And 86% treat AI output as a starting point rather than the final answer.

Stanford’s AI Index finds something similar from a productivity angle: measured AI gains are largest where work is structured, measurable, and relatively easy to evaluate, with smaller gains on work requiring deeper reasoning. Which sounds awfully familiar.

AI is very good at helping us make. The interesting part is deciding what deserves to be made.

Speed is everything. Direction is more expensive.

There's an old martial-arts saying—that you shouldn't fear the person who has practiced ten thousand kicks once. You should fear the person who has practiced one kick ten thousand times. I never took martial arts, so take what I say with a grain of salt, but I think you get the gist.

AI gives designers a lot more kicks. That's good. Make the prototype. Make three. Throw one away. Test another. Use a napkin sketch when a napkin sketch will do, and build the functioning version when feeling the interaction actually matters. Run the accessibility audit. Explore the management experience even though nobody asked for it yet. Then form an opinion.

Because the designer who can make something three times faster is absolutely somebody I'd pay attention to. But speed isn't valuable because it lets that designer ship three times as many rectangles. It's valuable because they can provoke more conversations, explore more dimensions of the problem, accumulate evidence faster, and arrive at a stronger conviction while everyone else is still arguing about the PRD.

The pixels were never the point. The decision was.