When AI Designs: The Case for Craft

I was in an all-hands the other day when a Product Manager presented a series of prototypes he'd generated in a few hours. Not concepts. Not wireframes. Working prototypes. They pulled from existing styles, handled multiple user flows, and looked polished enough that people in the room were already practically talking about shipping timelines. The reaction was enthusiastic. The reaction from designers was more complicated.

Because the prototypes were impressive, they were also wrong in ways that only a trained eye would catch. Pattern inconsistencies that would break at scale. Flows that worked for the demo but ignored how real users actually move through the product. An information hierarchy that looked clean with placeholder data and would fall apart the moment someone had more than three items in a list. The prototypes were a confident first draft wearing the clothes of a finished product.

There's a gap between what AI builds and what users actually need. It's not a gap of effort or intent. The Product Manager who built those prototypes is smart and cares about the product. It's a gap of craft: the accumulated understanding of how people interact with software, where patterns break, when something that tests well in a demo fails in real life. AI can close the distance between an idea and something that looks like a product. It can't close the distance between something that looks like a product and something that genuinely serves the person using it.

If you're a designer, this gap should feel familiar. It's where we've always lived. And it's getting more consequential, not less.

AI can close the distance between an idea and something that looks like a product. It can't close the distance between something that looks like a product and something that genuinely serves the person using it.

The Prototype Looks Done. It Isn't.

This is happening everywhere, and the pattern is consistent. An AI-generated prototype looks complete enough to feel decided. A stakeholder sees a working flow, recognizes the components, and their brain fills in the rest with optimism. They assume the edge cases have been handled because the happy path looks smooth. It's the product equivalent of a restaurant that nails the lighting and the menu typography but serves mediocre food. The presentation creates an expectation that the substance doesn't meet.

The gap between what AI builds and what users need isn't a failure of the technology. It's a category of problems that takes experience to see and skill to solve. And AI has made it harder to notice because the starting point looks so much more polished than it used to. When a wireframe looked like a wireframe, everyone understood it was a draft. When a prototype looks like a product, the team treats it like one. The conversation jumps from "what are we trying to learn?" to "when can we ship?" and that skip is where the cost of craft starts compounding.

Cheap Building, Expensive Judgment

The cost of building is dropping fast. Figma Make generates interfaces from design libraries. Cursor turns conversations into code. Claude Code can produce a functional prototype from a description. According to Figma's 2025 AI Report (opens in new tab), one in three product builders is now shipping AI-powered products (a 50% increase from the prior year), and 52% of those builders said design is more important for AI products than traditional ones.

Read that again. The people building with these tools are saying they need more design, not less.

A Productboard conversation with Intercom (opens in new tab) earlier this year framed the tension well: as the cost of building drops, the responsibility to choose well increases. Velocity stops being the differentiator. Judgment is. That tracks with what I'm seeing. The question has shifted from "can we build this?" to "should we build this?" and "is this the right version of this?" Those are design questions. Always have been.

Ethan Eismann, the CDO at Nubank (formerly SVP Design at Slack, and before that in leadership at Airbnb, Uber, Google, and Adobe), published a piece called "Embracing AI-First Design (opens in new tab)" in January that articulates this better than anyone I've read. He identifies four opportunities AI creates for designers: speed to learning, higher starting points, exposure of strategic value, and the increasing importance of taste. The through-line across all four is the same. When execution gets cheap, the value moves upstream. The people who can identify the right problems, evaluate solutions with a trained eye, and connect design decisions to real outcomes become more important, not less.

As Eismann writes (opens in new tab): "When anyone can produce competent work, competence stops being enough. The differentiation comes from having a point of view."

David Kossnick, Figma's Head of AI Products, put it differently at Config 2025 (opens in new tab): "All the human judgment, empathy, craft, taste is what it means to be the pilot, not the copilot." Designers aren't along for the ride. They're the ones who should be steering.

The Empathy Gap

Here's the pattern I keep seeing. Someone generates a prototype with AI. It looks plausible. The team starts optimizing around it before anyone has asked whether it solves the right problem. The gravitational pull toward "this looks done" is strong. The thing that looks built starts to feel decided.

Every designer has been in a meeting where a stakeholder fell in love with a mockup before the research was finished. That's not new. But AI compresses the timeline between "idea" and "looks like a product" to the point where the messy, necessary work of understanding user needs can get skipped entirely. Not maliciously. Just because the prototype is sitting right there, looking reasonable, and momentum is hard to argue with.

Alignment on desired user outcomes and the problems worth solving is underestimated as a UX discipline. It's not wireframes and flows. It's the ongoing work of making sure the team stays honest about who they're building for and why. As these tools evolve, we're going to need to apply our skills in empathy and vision to writing prompts that solve customer needs with real user-centric thinking, not just prompts that build features.

Which raises an interesting possibility: designers might be the best prompt writers in the building. Not because prompting is a deeply technical skill, but because the quality of your prompt is directly related to how well you understand the user, the constraints, and the intended outcome. That's been our job for decades. The medium is changing. The underlying ability is the same.

Designers might be the best prompt writers in the building. Not because prompting is a deeply technical skill, but because the quality of your prompt is directly related to how well you understand the user, the constraints, and the intended outcome.

We've Been Here Before

There's a pattern in how product organizations evolve, and it's worth remembering right now because it tells us where this is going.

There was a time when companies thought all they needed was engineers. Build the thing, ship the thing. Then the products got more complex and the roadmaps got messier, so product managers showed up to direct engineering efforts toward the right problems. Eventually, companies realized that functional wasn't the same as good, and designers became the competitive differentiator: the people who could take a working product and make it something people actually wanted to use.

AI is going to follow the same arc. Right now, the conversation is mostly about Product Managers and engineers using AI to build faster. The prototypes look fine. The outputs are functional. And for a while, that will be enough. But the pattern inconsistencies will pile up. The products that were "fast to build" will become "expensive to fix." The gap between "this technically works" and "this actually serves the user" will become impossible to ignore.

At that point, the industry will rediscover that it needs designers in the process. The question is whether that rediscovery happens after a bunch of poorly crafted AI products hit the market, or whether designers figure out how to provide differentiated value before the slop gets out of hand. I'd rather we be proactive than wait for the usual cycle of build-it-fast, wonder-why-it's-bad, hire-designers-to-fix-it.

The difference this time is speed. Previous transitions played out over years. This one is moving in months. That compresses the window for designers to establish their position.

Systems over Heroes

Eismann makes an argument in "Embracing AI-First Design (opens in new tab)" that I think is the most practical thing I've read on this topic. He writes that great outcomes depending on who happens to be in the room are fragile. They can't scale. The goal is to build systems that make strong design the default.

Think about that in the context of AI-generated work. If your organization can define what good looks like and encode it into tools, workflows, and review processes, AI becomes a force multiplier. You can scale quality in ways that weren't possible before. If you can't do that, AI produces inconsistency at scale. The same tool that helps good designers move faster helps everyone else flood the product with mediocrity.

The question, as Eismann frames it, isn't "how do we use AI?" It's "do we have the systems to make AI-assisted design reliably excellent?"

That reframe matters. It shifts the conversation from defensive ("we need to protect craft") to proactive ("we need to build the infrastructure that makes quality the default, with or without AI"). And the work of building that infrastructure is design work. It requires understanding quality deeply enough to articulate it. It requires thinking in patterns and principles, not just individual solutions. Design systems, quality standards, governance, codified craft. This is where the leverage is.

David Robinson makes a related point (opens in new tab) about design systems becoming infrastructure for AI, not just for designers. If your tokens aren't machine-readable, if your component logic isn't documented in a way an LLM can parse, you're creating friction for every AI workflow your team builds. The design system is becoming the intelligence layer that keeps AI outputs on-brand and on-quality. That's a practical investment that pays off immediately.

What Design Teams Need to Do

I don't think any of this is fully figured out yet. But there are a handful of things I think design teams need to get serious about, and soon.

Pull Information Forward

The quality of any AI output is only as good as the inputs, and right now too much of that input lives in silos. Your Product Manager has the business case, the competitive analysis, and the market opportunity data. Engineering knows the technical constraints and the infrastructure realities that will shape what's actually buildable. Research has the user pain points, the behavioral patterns, and the jobs-to-be-done. If any of that context is sitting in a doc, a deck, or someone's head when a designer starts prompting an AI tool, the output is going to reflect those gaps.

Designers need to actively pull this information from their cross-functional partners early, before the first prompt gets written. That means asking Product Managers for the business goals and competitive context upfront, not after a prototype exists. It means understanding from engineering what technical constraints or dependencies will shape the solution. It means making sure research insights and user data are part of the brief, not an afterthought. The designer's job has always been to synthesize across these inputs to deliver something that's both viable and genuinely useful. AI doesn't change that. It just makes the cost of missing an input higher, because the gap between a prototype built with full context and one built without it is the gap between a solution and a guess that happens to look like one.

Compress Timelines without Compressing Quality

Months of work need to become weeks. Weeks need to become days. That doesn't mean cutting corners. It means identifying which parts of our current process are genuinely valuable and which are artifacts of older workflows that AI can now absorb. Think of it like cooking: mise en place matters, but if you're still hand-grinding your spices when there's a perfectly good tool that does it in seconds, you're confusing ritual with craft. The triage requires knowing your practice well enough to distinguish what's essential from what's habitual.

Learn the New Tools like We Learned the Old Ones

We had to learn Sketch, then Figma. We had to learn prototyping, then design systems. This is the same thing. We have to learn how to use the right AI tools and workflows to communicate the level of craft and ease-of-use design should deliver. The tool literacy is table stakes. The craft on top of it is what matters.

Eismann frames this (opens in new tab) as developing "curatorial judgment," the shift from "can I make this?" to "do I know excellent when I see it?" That distinction is worth sitting with. Without strong creative direction, AI output converges toward the generic. The same tools trained on the same data produce the same median outcomes. Differentiation comes from knowing what you want and having the taste to recognize when you've found it.

Codify What Good Looks Like

The more execution is automated, the more you need explicit standards. Comprehensive Design systems. Craft scorecards. Design quality indices. Clear criteria that teams can evaluate against. This is infrastructure work, not busy work. It pays off every time someone uses AI to generate design work, because it gives the team a shared definition of "done" that goes beyond "it runs."

Shift How We Communicate Value

This might be the hardest one. It's tough to beat pure speed, so we have to articulate business and customer value over "taste" and "consistency." Non-designers can look at AI output and call it tasteful enough. They can call it consistent enough. The argument for design can't rest on subjective quality assessments anymore. It has to connect to measurable outcomes: retention, task success, reduced support tickets, NPS, conversion. Taste is real, but it needs to be backed up by data if we want it taken seriously in a room full of people who just watched someone build a prototype in twenty minutes.

Cheap Building Makes Design More Important

Here's where I land on this.

The instinct among some designers is to view AI as a threat to the profession. I understand the impulse, but I think it's backwards. AI doesn't diminish the need for design. It intensifies it. When the floor for "acceptable" output rises, the bar for what's actually good rises with it. When building is cheap, the premium moves to the people who know what's worth building and can tell the difference between something that works and something that works well.

Eismann calls design the soul of the product (opens in new tab), "what binds together how something works, how it feels, and how it fits into someone's life." AI doesn't threaten that. It makes it more visible. When everyone has access to the same generative tools, the ability to bring judgment, vision, and a genuine understanding of users to the work becomes the differentiator. Not the tools. Not the speed. The thinking.

Modus Create's research (opens in new tab) shows that 75% of product leaders still struggle to follow through on strategic elements of product development, even with AI accelerating execution. Fast shipping without clear strategy just means you arrive at the wrong destination sooner. That's where design leadership matters: keeping the team oriented around problems worth solving, not just solutions that are easy to build.

A twenty-minute prototype is impressive. It's also incomplete. And the distance between impressive and excellent is where designers live. That distance isn't getting smaller. It's getting more consequential.

A twenty-minute prototype is impressive. It's also incomplete. And the distance between impressive and excellent is where designers live.

The work ahead of us is better than the work behind us, but only if we're honest about what's changing and willing to adapt how we operate. AI is not the end of craft. It's the reason craft matters more than it has in years.