Speed is not a product strategy

AI design tools have collapsed timelines that once seemed fixed.

Roger described a traditional product design process in which teams might speak with customers, disappear for a month or two, and eventually return with a Figma prototype. Today, AI prototyping can produce something testable in a week or even less. Teams can show an idea to a customer, revise it after the conversation and bring an improved version within hours.

That is a meaningful advance. The distance between hypothesis and evidence is shrinking.

But speed carries a seductive logic: if making something is faster, making more things must be better. Wrong. 

Roger sees the danger in that assumption. Earlier this year, he noted, parts of Silicon Valley became enamored with measuring AI usage itself—the number of tokens consumed, for instance. Such metrics reward activity rather than accomplishment.

“You have to be measuring and incentivizing the right things,” Roger said. “It’s not about outcomes” when the focus is simply on volume.

The better model, he argued, is familiar to anyone who has practiced disciplined product development: build something, release it, measure what happens and repeat. “It’s really the scientific method,” he said. “And we can’t forget that.”

AI-powered product development can make that cycle dramatically faster. It cannot decide whether the hypothesis was worth testing in the first place.

The best AI ideas may begin without AI

One of the most revealing stories from the conversation involves no prompt engineering at all.

BuildOps makes software for commercial contractors, serving people ranging from CFOs and dispatchers to HVAC, electrical and plumbing technicians working in the field. Earlier this year, a BuildOps designer went out to meet technicians where they actually worked.

She discovered something mundane and consequential.

At the end of a long day—sometimes spent sweating in a boiler room or working on a rooftop—technicians could spend another 10 to 20 minutes sitting in their trucks typing notes about the job they had just completed. Tired workers sometimes produced terse or incomplete records. Those gaps could later contribute to billing disputes or leave another technician without sufficient service history.

The team also uncovered a smaller detail with larger implications: some Android devices didn’t offer the same easily accessible voice-dictation experience the designers, many of them iPhone users, took for granted.

No language model would have noticed the heat on the roof, the exhaustion after nine hours of physical labor or the phone in a technician’s hand.

From that customer research came an AI feature called Visit Summaries. Instead of typing everything, technicians can press a prominent microphone button and talk. AI cleans up the account and incorporates relevant context about the property, equipment, parts used, and so forth.

The sequence matters. The team did not begin with AI and search for somewhere to wedge it in. It began with observing a person experiencing friction and then finding a way AI could help.

That is human-centered design with AI, rather than AI looking for a human-centered use case.

Trust is earned before it is automated

The BuildOps example also exposes another problem with AI adoption: customers may be considerably less interested in AI than the people building it.

When Roger’s team spoke with technicians about ambitious AI capabilities, the initial response was decidedly practical.

“Just give us the basics first,” Roger recalled hearing. “Just make sure the app is reliable and it’s speedy and we can do what we do.”

Only after those needs were addressed did the conversation shift. When the team described an AI experience that could reduce tedious end-of-day typing, technicians saw the value.

The lesson for AI and customer trust is deceptively simple. Reliability comes before novelty.

“I think you have to earn that trust,” Roger said, by having a product that works reliably.

Product teams sometimes talk about trust as though it were another interface component: add transparency here, an explanation there, perhaps a confirmation screen before the model acts. Those mechanisms can matter. But trust is also cumulative. It is the memory of all the times a product did what it promised.

AI does not get to skip that process simply because it is impressive.

AI makes human judgment more valuable, not less

The paradox of AI in product design is that the easier it becomes to generate work, the more important it becomes to judge that work.

Roger saw this firsthand with AI-generated documents. Teams could suddenly create long product requirements documents at remarkable speed. The problem was that generating a document and thinking through a problem were not the same activity.

In some cases, people treated the output as “good enough,” leaving someone downstream—a designer or engineer—to discover the holes.

His team eventually established a straightforward expectation: if you generate something with AI, you own it.

“What it says there reflects you and your professionalism,” Roger said.

This may become one of the defining principles of AI-powered product development. Automation can transfer labor, but it cannot transfer accountability.

It also helps explain why Roger’s criteria for hiring designers have not radically changed. Yes, he expects candidates to have experimented with AI design tools such as Figma Make or similar products. But tool proficiency is not the differentiator.

What he looks for is craft and judgment: typography, layout, color, systems thinking, empathy, the ability to speak with users and the ability to turn qualitative information into actionable insights.

When everyone has access to powerful generation tools, knowing how to generate becomes less scarce. Knowing what is good becomes more scarce.

Leaders need to play with the material

There is another side to Roger’s argument. Defending design fundamentals cannot become an excuse for ignoring AI.

He thinks design leaders should get their hands dirty.

Roger described AI as a new “material” for designers, distinct from the paper and screens that defined previous generations of design work. Materials have properties and limitations that are difficult to understand abstractly. A carpenter learns about the different wood grain and how it cuts or takes a screw partly by working with it; a designer needs a similarly intimate understanding of AI.

Roger has experimented with agents, local language models, websites and applications. Before the San Diego Comic-Con, frustrated by its sprawling schedule, he spent a weekend using Claude Code to build a scheduling app for himself.

Not every experiment needs a business case. 

“And why do we fall, Bruce? So we can learn to pick ourselves up.”

— Thomas Wayne, Batman Begins

Sometimes the point is learning where the material bends and where it breaks. That understanding matters when an AI and design leadership challenge moves from experimentation to strategy.

“You can’t really lead a team through this period without understanding it intimately,” Roger said.

The uncomfortable question about the next generation

There is, however, a deeper problem waiting beneath all this newfound efficiency.

For decades, junior designers learned partly by doing work senior colleagues no longer wanted to do. The tasks could be repetitive and unglamorous, but repetition trained the eye and developed instincts.

AI can now absorb much of that work.

Roger worries about what happens next. He has written about what he calls a brewing design talent crisis: if companies no longer need junior employees for foundational tasks, where and how will future senior designers acquire their judgment?

He remembers learning old photo-mechanical production techniques in design school—skills he never directly needed professionally. Yet doing that work trained his eye and helped him understand the digital tools that followed.

Efficiency, in other words, can remove more than waste. It can remove practice.

That should give product and design leaders pause. The challenge is not to preserve tedious work for nostalgia’s sake. It is to identify what people were learning while doing it and create new ways to develop those abilities.

AI may shorten the road to an artifact. It cannot eliminate the road to expertise.

And that brings the conversation back to where Roger began. AI prototyping, customer research tools and generative AI for designers will continue to evolve. The interfaces will change. The workflows will change. Some jobs almost certainly will, too.

But the essential obligation of product design remains stubbornly human: observe carefully, think critically and exercise judgment about what should exist.

As Roger put it, “We should lean into our superpowers that way.”

Additional resources

• Roger Wong on LinkedIn
• Roger's website: RogerWong.me
• Roger newsletter
• Jason Giles on LinkedIn

• The AI-native product loop: build, test, learn—without slowing down: This on-demand webinar explores how AI-native product teams can integrate real customer feedback into fast development cycles so validation becomes part of the workflow rather than a separate step.

• How to build customer-centric products without slowing down development: This guide addresses the tension at the center of Roger’s argument: teams want greater speed, but speed without customer validation can result in wasted effort and poor outcomes. It covers rapid prototype validation, continuous customer testing, AI-assisted insights, and collaboration across product, UX, and engineering.

• UX for AI: why building faster is no substitute for building the right thing: This Insights Unlocked episode with Greg Nudelman distinguishes between using AI to accelerate UX work and designing AI experiences themselves, while emphasizing customer discovery, iterative validation, human oversight, risk, and measuring real-world outcomes.

• AI design decision making: how to build confidence under pressure: This blog post addresses the “confidence gap” created when AI allows design teams to prototype and iterate faster than ever. Its focus on balancing speed with human insight, accountability, and defensible design decisions closely complements Roger’s point that AI-generated output still requires human ownership and judgment.

Frequently asked questions (FAQs) about AI in product design