The Role of Design Constraints for AI Solutions
It’s easy to buy into the hype and hysteria surrounding AI and what it can spit out for you. AI is changing our work as designers. Does it mean we will do less UI work? Perhaps. Though a Design System already takes (or will) take care of many of these decisions. No one needs to design their own button.
Yet these “advances” miss the point. Designers are not hired to create a great looking UI. The main value designers bring to a company is “visualizing concrete solutions that serve human needs and goals within certain constraints” (Goodwin, p. 3). Sure, Claude or Figma Make can create something faster, but is it the right thing? Does it solve a real human need? Does it respect the constraints of the environment that it lives in? That’s the work of a Designer.
At this point the Aidvocates nod, “Aha, yes, that is where designers ought to be—prompting their AI junior design partner. Guiding it here and there to produce a finished product.” And that’s true. In May of 2026, designers have the ability to render convincing replicas of websites, applications, and landing pages using little more than a PRD written by another LLM and supervised by a senior product manager. The results have been astounding. But do they solve a problem real humans have?
In 2017, I attended a workshop led by J. Paul Neeley and Elliott P. Montgomery on Speculative Design.1 I sat next to Chris Noessel, then of Cooper Design. I recall a somewhat heated exchange between them on the efficacy of a particular example. The item was a new health device located in a doctor’s office. Research had discovered a certain species of bee reacted to the breath of cancer patients. A designer leveraged this information to create a speculative product. The device itself was a large, glass enclosure with a bee inside. The glass had a small opening for the patient to breathe into, activating the bee. By watching the bee’s behavior, the doctor of the future would be able to discern if the patient had cancer.
Chris was beside himself. Drawing on years of experience designing products for doctors, he picked apart flaws related to the bauble. It was breakable glass, it was heavy, there was no room for such a device in any doctor’s office. In short, he knew the environment the object would live in. The designer had understood none of it. Neeley replied that the purpose for the piece, and speculative design in general, is a discussion not the object itself.
Julian Bleecker at Near Future Labs provides an excellent synthesis of these perspectives. He creates plausible near future artifacts in everyday environments. Like speculative futures, artifacts are provocative, yet situated in a real world. Cereal boxes of Cricket Crunch. A food truck owned and operated by an autonomous AI. A SkyMall-esque catalogue of items from some near future. They all live in the “future mundane.”
AI makes the future easy to depict. We can spin up a product as soon as we dream it. With a little more prompting, it takes shape and we have a reasonable prototype to test our new idea. Yet starting here risks building the wrong thing. Designing without understanding users and their context leads to products that fail them.
Achieving relevant products means spending time with the people you serve. Study their environment. Question to find out their end goals. Sympathize with their fears and frustrations. Research the other tools they use today. Learn about what holds them back, understand the constraints and why they’re there. In his new book, Inside the Box, David Epstein discusses Bent Flyvbjerg and his database of large projects. Most fail to deliver on time and on budget. Even fewer achieve their intended purpose. The biggest success factor was the teams’ approach. “They [the successful ones] take the opposite course: ‘Think slow, act fast.’ They take time to work out boundaries that make the scope of a project clear. Then, when it’s time to hit the gas, work can proceed quickly” (p. 14).
Recently, someone on my team sent a write up of Claude Design’s abilities in creating a coffee delivery app. The breathless report made me wonder if it was satire. “Not a wireframe. Not a mockup. A working prototype — in under 10 minutes.” Wow. “Then someone said it out loud — damn, is this the end of designers?” Eye roll. In the end, it was a coffee ordering app. To get this far, they downloaded a free coffee-themed template from the Figma Community and let Claude have a go at it. The app itself is a very basic PMP to cart experience. There’s just one thing missing from the equation. There’s no one to buy it. The app sold no coffee, the team only made a cute MVP that looks like every other coffee app out there. Could it work? Maybe. But to know, you have to know the market, the customers, why they’re not satisfied with their current coffee ordering solution. Doing that requires the real work of Design—Research, Modeling, Defining requirements, Creating the Design Framework, and Design refinement. AI design tools can assist, but they don’t invent what people need.
Design is in danger of losing the plot. We’ve bought into the idea everyone wants to sign up for yet another digital service. That this app will definitely make their life easier and it’s ok if it is gone in a year. All that’s needed is to get that product to market. Sure there will be issues. Unfortunate events like this will cease as models improve. Yet that misses the point. These catastrophes are a part of the learning that our customers mete out. You are the price of progress.
There’s a better way. It starts by remembering that our role is to help users achieve their goals and overcome the obstacles that get in their way. This isn’t the work of AI, it’s our work. As AI matures, it will assist us more in that goal. Your favorite AI is a product itself. Its job is to help you achieve your goals and overcome obstacles. Hold it to that and demand excellence. Those we serve have the right to nothing less.
Bibliography
Arnis, A. (2026, April 21). Design was never the bottleneck. Our Collective Futures. https://collectivefutures.blog/design-was-never-the-bottleneck/
Epstein, D. (2026). Inside the Box: How Constraints Make Us Better (First) [Hard cover]. Riverhead Books.
Goodwin, K. (2009). Designing for the digital age: how to create human-centered products and services. Wiley Pub.
Lopatto, E. (2026, April 20). Silicon Valley has forgotten what normal people want. The Verge. https://www.theverge.com/tldr/915176/nft-metaverse-ai-weirdos
Neeley, J. P., & Montgomery, E. P. (2017, February 5). Speculative Design: Futures Prototyping for Research and Strategy. Interaction 17, New York, NY. https://interaction17.ixda.org/session/speculative-design-futures-prototyping-research-strategy/index.html
Samsonov, P. (n.d.). Software is a coordination problem. AI can’t help you with that. The Product Picnic. Retrieved May 7, 2026, from https://productpicnic.beehiiv.com/p/software-is-a-coordination-problem-ai-can-t-help-you-with-that
Tyson, M. (2026, April 27). Claude-powered AI coding agent deletes entire company database in 9 seconds—Backups zapped, after Cursor tool powered by Anthropic’s Claude goes rogue. Tom’s Hardware. https://www.tomshardware.com/tech-industry/artificial-intelligence/claude-powered-ai-coding-agent-deletes-entire-company-database-in-9-seconds-backups-zapped-after-cursor-tool-powered-by-anthropics-claude-goes-rogue
- J. Paul Neeley gave this presentation in a number of venues. Here is one he gave at the Push Conference in December 2017. ↩︎
