Looking at Fundmates now, I see the beginning of my shift towards working with AI. My portfolio already has two case studies about this project. They show interfaces and describe product problems. Another part stayed behind the scenes: how I tried to understand what was going on and gradually brought neural networks into my work.

Fundmates was a difficult context for me. I had tasks from the product owner, layouts, individual screens, and references. I could not always understand it all at once and turn it into a coherent user journey.

I began working with ChatGPT and Claude in parallel. I gave them interface screenshots, my designs, examples of other solutions, and descriptions of what needed doing. I wanted to assemble a reasonably clear flow: what happens to the user, in what order, and how that can appear on screen.

First I had to understand the task myself

A case study usually shows an assembled result. It has a logic, the screens are in the right order, and everything looks as though you knew where you were going from the start.

There were many moments in this project when I did not yet understand how to assemble the interface. That was exactly the state of the task I brought to the models. I supplied complex context and got back an explanation or sequence of actions I could work with.

Then I returned to Figma and implemented the scenario in the designs. AI helped me understand things and organise the logic, while I still designed the interface with familiar design tools.

So I would not describe Fundmates today as a product I developed entirely with neural networks. It was a gradual change in my workflow. ChatGPT and Claude were already participating, although the final work still largely looked like a conventional design project.

First attempts in Figma Make

To test hypotheses, I began making early versions in Figma Make. Their UI was poor. They were far from the visual result I wanted.

But they communicated ideas well. I could assemble a version of a scenario and see how the hypothesis worked. At that point, it was useful to me: even a visually weak prototype helped demonstrate what I was trying to devise.

The Figma designs and those early Make versions had different jobs. In the designs, I built the interface. In Make, I tried to bring an idea to life and test the interaction sequence. It was not yet a finished application you could simply hand to a user.

Now I want to show both parts in the case study. Not just selected polished screens, but the attempts that reveal the search itself. Without them, an important part of the Fundmates story and my own way of working disappears.

From that experience to today’s projects

Today I use AI to develop websites and digital products. But the transition started earlier, with a specific need: understanding a complex task and working out what interface should come from it.

Fundmates established a sequence I still recognise: gather context, discuss it with a model, get a clear scenario, build a version, and examine the result. Back then, I completed much of that sequence manually in Figma. Later, more working code entered it.

For me, Fundmates records a transition in the design process. I gave ChatGPT and Claude complex context, received a clear flow, and implemented it in Figma. Early Make helped show hypotheses, although its UI fell short of the designs. I do not call this a product developed entirely by neural networks.

Related stories: Yango Pay prototypes · Pilates.me on Framer. An AI web application for the Fundmates startup. Sell One Video: turning a video’s future revenue into money today.