Experiment like never before
#68: How AI enables the creation and testing of innovative solutions
Experiment like never before
Recently, I took part in an internal workshop where we worked on a new registration flow for one of our products. At one point, someone from the team asked, “What if we approached registration in a completely new way and got rid of the typical sign-up form altogether?” The idea sounded a bit abstract at first, but it quickly caught on, and we spent some time discussing it. Meanwhile, another team member created a fully functional prototype in just a few minutes using Figma Make, something we could instantly validate with users.
Not long ago, this would have required building clickable mockups that probably wouldn’t fully convey our idea and would have been a weak basis for validation, or involving developers to build a prototype we’d likely have to wait several days for. Today, we can have it almost instantly. Not to mention, we probably wouldn’t even be able to allocate resources to create something that abstract.
Thanks to tools like Figma Make, we can very quickly build and test prototypes of the solutions we are currently working on (as we did with our new registration flow), as well as those that might initially seem completely crazy. This opens up more room than ever before for experimentation and testing entirely new approaches in product design.
Not just crazy ideas
Even though at this stage we do not have much room to experiment with completely wild ideas, AI support is still invaluable for more everyday tasks where having a prototype is essential to ultimately deliver a product of the right quality.
The biggest value I see in the easier creation of prototypes, whether from scratch or using previously designed screens, lies in two areas: rapid prototyping, which we use most often, and the research through design method.
Rapid prototyping
Rapid prototyping is about quickly creating a prototype, collecting feedback, and improving the design. Time plays a crucial role, as the method assumes that the entire process happens relatively fast. Until recently, this was a major limitation because it meant we could not build complex, high-fidelity prototypes and had to rely on low- or mid-fidelity wireframes to gather feedback quickly and move forward.
Now we can create high-quality prototypes much faster, ones that are not very different from standard working solutions. This allows us to gather more valuable feedback than with low-fidelity wireframes and to test a greater number of ideas, which gives us more room for experimentation.
The value of faster and higher-quality prototyping is also visible in more traditional design and testing processes with users, which typically took much longer than rapid prototyping. In addition, AI supports us in preparing research scenarios, analyzing results, and summarizing findings, which further reduces the time needed and lets us either move forward with the project faster or accomplish more in the same timeframe.
With the current level of AI support and the significant reduction in the time required to prepare prototypes and tests, the line between rapid prototyping and a more traditional process is becoming increasingly blurred. Of course, time is not the only difference between these two approaches. Factors such as scope, goals, and sample size also matter. However, distinguishing between quick and non-quick usability tests no longer makes much sense, since the time needed to build prototypes, run tests, and analyze results is significantly shortened in both cases. This allows us to focus on the goals we want to achieve through testing without the compromises that used to come from time constraints.
In our case, while working on a registration form, the support of Figma Make proved incredibly useful. From all the working versions, we prepared two final ones that we wanted to test with users to find out which approach would result in more users completing the registration process. After designing the wireframes, we transferred them to Make to create fully functional prototypes. The entire form behaved exactly as it would in production. All fields were interactive, validation worked, we could simulate SMS code confirmations, and even check how users understood password strength. The form knew the password requirements and displayed information about whether the entered password met them and how secure it was. What would normally take a developer several days to build was achieved, including prototype refinements, in less than half a day. We also did not have to rely on traditional wireframe-based prototypes, which are not well suited for testing forms.
Importantly, we did not use AI tools to generate static screens. Those were still designed manually, and the prototypes were based on them, as the results we obtained when experimenting with AI-generated sketches and ideas were not satisfactory.
Research through design
Research through design is a method that uses design as a way to conduct research and generate new knowledge. In this approach, we create prototypes that serve as the foundation for our studies. Even during the process of creating them, designers can already draw valuable insights. RtD supports gaining knowledge that would be difficult or even impossible to obtain without using previously created prototypes in research, both on the side of designers and the users participating in the studies.
From my observations, the research through design method is not as popular as other research approaches. Yet, this is precisely where, if the organization provides the space, we can experiment and seek innovation. It allows us to test and gather knowledge about completely new, previously unexplored solutions and directions.
However, in many organizations, there is often no room for experimentation or innovation. It is a more time-consuming process and may not fit into current priorities. Thanks to AI, this might be changing.
I have already mentioned the role of AI in supporting prototyping. It now enables the rapid creation of advanced prototypes that are almost indistinguishable from real products. In the case of RtD, the value gained from this is equally significant, as it opens up new opportunities for product teams to explore ideas that previously did not fit into their schedules. Experimentation and innovation have become faster and easier, which means they now require fewer resources and can be more easily integrated into ongoing work.
RtD the new product discovery?
As I mentioned earlier, this method is still not widely adopted, but that may soon change. From conversations with other designers at a recent conference, I learned that many product companies are testing approaches inspired by RtD. Instead of conducting time-consuming discovery processes, they shorten them by moving directly to creating a prototype and learning from it.
Thanks to AI, designers can now build and test prototypes much faster than a traditional discovery process would allow. Could an RtD-inspired approach become the new standard for product discovery, offering equal (or even greater) value in less time? It is probably too early to say, but it is certainly something worth keeping an eye on.
A few thoughts to wrap up
To conclude, I want to share a few reflections from our recent experiences. These are not meant to be in-depth tips on using AI tools (we will cover that in a separate article), but rather a few simple suggestions to apply when creating prototypes. For those who are already at an advanced level, some of these points may seem obvious.
First, tools such as Figma Make and similar ones, while improving rapidly, are still far from perfect and are not suitable for every use case. Based on our experience, we are currently unable to achieve high-quality UI output or accurate implementation of our design system. For now, we use these tools primarily for prototyping and testing purposes.
It is also crucial to prepare prompts carefully and treat the tool more like a junior designer who needs clear guidance rather than a senior designer who will figure everything out on their own. The more context and clear instructions we provide, the better the outcome. Editing generated artifacts can also be tricky, as requesting changes sometimes alters more than we intend.
That said, especially in the early stages, it is worth testing different tools to see which ones fit your needs and limitations best. For example, one of our current constraints is the decision not to sync our Storybook library with AI tools. Keep an eye on updates too, because with the current pace of development, entirely new capabilities can appear overnight. I highly recommend following official sources such as OpenAI or Anthropic and their team members rather than self-proclaimed experts trying to sell you courses.
Support our work
If you enjoy our newsletter, consider switching to the Premium plan to show your support! For just $5 a month (or $45 a year), you will get full access to our archives and free access to ebooks from Fundament Library.
Partnerships and socials
Partner with us via Passionfroot
Mateusz’s LinkedIn