I Was Already Researching AI in Design in 2023. Here’s What Changed.

In 2023 I wrote a thesis on AI in the design process. Three years later, almost everything has changed. Except the most important part.

In 2023, I co-authored a thesis on AI tools in the concept phase of design. The research question was simple: how are AI tools used in the early stages of the design process to create usable interfaces? Three years later, the AI in design process landscape looks almost unrecognizable. But the core insight from that research still holds.

The thesis: AI tools in the concept phase

The paper, Skapa användarcentrerad design med AI-verktyg i konceptfasen, was a qualitative literature study analyzing 32 research articles through a framework called brukskvalitetsprismat (the usability quality prism). We looked at AI tools across six dimensions: technical, practical, communicative, organizational, aesthetic, and ethical.

Our conclusion: AI tools have significant potential to improve the design process and create user-centered interfaces, but they should complement human creativity, not replace it. That was the finding in 2023. It turned out to be more true than we expected.

You can read the full thesis on DiVA.

What we got right

The thesis identified that the concept phase is where designers spend the most time, and where AI could have the biggest impact. Research at the time showed that 60% of a designer’s time went to analyzing information and exploring unstructured resources. We argued that AI tools could dramatically reduce that time.

That prediction landed. In 2026, tools like Figma Make, Google Stitch, and Flowstep generate complete UI flows from text prompts. What used to take hours of wireframing now happens in minutes. Teams using AI tools are shipping features 40 to 60 percent faster than those still working manually.

We also emphasized the ethical dimension. We wrote about the need for transparency in AI-based design decisions, the risks of unintentional discrimination in training data, and the importance of protecting user privacy. These concerns have only intensified. Today, every serious conversation about AI design tools includes questions about data handling, bias, and informed consent.

What surprised me

The speed of change. In 2023, we analyzed tools like Miro, UserTesting, Figma with PromptInfuser, and GPT-4. PromptInfuser was a research plugin that let designers run LLM prompts inside Figma mockups. It felt experimental, almost academic.

Today, that same concept is native to multiple platforms. Figma Make pulls in your existing design system components so AI-generated prototypes match your brand from the start. Google acquired Galileo AI and relaunched it as Stitch, a free tool that generates Figma-ready designs from prompts, images, or sketches. Tools like Moonchild AI create both high and low-fidelity designs from conversational prompts.

The thesis talked about NLP and machine learning as enabling technologies. We were right about the importance of those technologies, but we underestimated how quickly they would become invisible. Designers today don’t think about NLP when they use Stitch. They just describe what they need and get a result.

What I got wrong (or at least, incomplete)

Our framework focused heavily on individual tools and their capabilities. What I didn’t fully appreciate was how much the workflow itself would change. It’s not just that better tools exist. It’s that the entire relationship between concept, prototype, and production has compressed.

In 2023, the concept phase, prototyping phase, and development phase were distinct stages with clear handoffs. In 2026, the boundaries are blurring. AI tools generate production-ready code alongside design mockups. The concept phase and the build phase are starting to merge.

That’s a bigger shift than just having faster wireframing tools. It changes what it means to be a designer.

The one thing that hasn’t changed

AI still can’t tell you which problem to solve.

The concept phase isn’t just about generating screens. It’s about understanding users, identifying real needs, making judgment calls about what matters. That requires empathy, context, and the kind of thinking that comes from actually knowing the people you’re designing for.

Our thesis concluded that AI tools should be a complement to human creativity and insight, not a replacement. Three years of rapid tool development haven’t changed that. If anything, the human side of the design process matters more now, because the technical barriers to producing screens have largely disappeared. The differentiator is knowing which screens to produce and why.

That’s what I’ll dig into in the next post in this series: a practical framework for evaluating whether an AI tool actually helps you design better, or just helps you design faster.