Using AI to Transform UI/UX Workflows: A Real-World Case Study
I was recently tasked by a group of sales team members to design and develop a B2B Account Dashboard for their select clients. The app would enable key point of contacts like HR and Learning and Development professionals on the client side to log in to the app, enroll their executives in programs, and track their enrollments through the app.
Sounds uncomplicated. I can breeze through this and scope it reasonably easily. So, I conducted several interviews with business stakeholders to gather their feedback and ultimately developed the product requirement document (PRD) for the app. While building the PRD, I worked in Salesforce and Figma to capture screenshots to include in the document. These visuals are incredibly helpful in conveying the message to multiple teams.
As I typed the document, pasted the screenshots, and added descriptions of the screenshots, I carefully described the UI/UX. I thought to myself, 'Wait a minute, can I feed some of this to ChatGPT and ask it to create the screenshots for me?'
Well, of course I could! Ran to ChatGPT, started typing things like, “create an image of a table with five columns, first column is for the name, second column is for the email”, and so on. Sure enough, AI was producing reasonably good images that I started adding to my PRD. I was so proud of myself, I was going to save so many hours of work!
A few weeks later, I was in a meeting with one of the developers I work with, and we were discussing the app for them to timeline it for me. I was reviewing the PRD and shared my 'aha!' moment, thinking I was so smart, saving a lot of time, and being very efficient.
They didn’t hesitate to stop me on my tracks. One of them said, Hey, Cem, have you heard of tools like Replit, Loveable, etc.?
I said, of course, I even tried them out last year; they were unreliable, with results that were hit or miss, and they made me feel frustrated, so I gave up on them.
They were quick to correct me; one of them said that the tools have come a long way since last year. They are more reliable, and the results are promising, especially if you are looking to get quick feedback on an app or feature without committing to complete development.
This was sometime in June 2025 when I said I would give it another spin. So off I went back to try Replit again.
This time, I took all my PRD as a whole and pasted it into Replit’s prompt. Then I waited; it came back telling me things about the development and design approach. I said, OK, Replit, do your thing. About 20 minutes passed, and it presented its first result. I was like, hmmm, this thing actually works now?! I was intrigued and immediately noticed that the result I had obtained a year ago was significantly different; now it was producing results that could actually work, at least for my purposes.
As I mentioned, I frequently interview business teams and rely on quick mock-ups to confirm needs or convey an idea during the process. I’ve relied on design tools like Figma, Adobe XD, Lucid Chart, hand sketches etc. to create these diagrams, mockups, and UX schemas, all this time. But now Replit was offering even more tangible.
With natural language instructions and previously created PRDs, Replit was able to make clickable, mobile-friendly prototypes within minutes. This process used to take me days of work to put together with the old method.
I immediately put this new method to work. The first was the B2B Account Dashboard; within a day, I had a very complex web app prototyped, without a single developer or designer. I sent a link to the Sales and Marketing teams to the Replit site, and asked for their feedback. Of course, obtaining the input now is the part that takes longer, as it involves waiting days to receive it from one team, and the other team hasn’t even had a chance to reply.
As a second trial, I took on a request from a marketing team to develop a pop-up banner on their website, again, fed the PRD to Replit, and within minutes a prototype of a pop-up was presented, with a control panel that moved it to different positions on the whole web page, with CTA buttons and images, and themes to choose from. Within the next hour, the prototype was reviewed by the team, and by the end of the day, a consensus was reached on positioning and styling, which was then passed on to the developers.
To reiterate, in the development phase, steps that involve human review, such as interviews and gathering feedback, remain unaffected by AI technology, so far. However, the designing and prototyping have significantly improved in terms of delivery timeline, accuracy, and malleability of the produced prototype.
If you are interested in learning more about how AI can help you prototype faster, please don't hesitate to reach out to me. I am happy to talk.
