As a UI designer, I have seen AI tools rapidly enter conversations about design workflows, but I had not had the chance to fully test their capabilities in a real project. Until now.
The goal was simple: explore Figma Make, with comparisons to Magic Patterns and Google Stitch, to see what these AI-powered tools could achieve in a real-world design scenario. Could AI take a vague concept and turn it into a clickable prototype, reducing the hands-on design time I would typically need?
I decided to test this on a prototype for a concept tool called Self Serve – an imagined platform designed to simplify the process of connecting clients with creative experts through the use of AI.
The aim was to understand how far modern AI design tools could go in creating usable UI prototypes with minimal manual input. By comparing tools like Figma Make, Magic Patterns, and Google Stitch under the same brief, I wanted to evaluate their strengths, weaknesses, usability, and potential impact on a designer’s workflow.
Setting the scene: what is 'Self Serve'?
The concept behind Self Serve is straightforward:
- Clients login and upload or write a brief.
- AI summarises the brief, allowing for quick edits or keyword additions.
- The AI matches the brief with relevant creative experts.
- Clients review profiles, shortlist their favourites, and send emails automatically.
This approach shifts administrative effort away from internal teams and puts more control directly into the client’s hands, with AI handling the logic in the background.
Choosing the tools and defining the process
To make the comparison fair, I worked with ChatGPT to develop a clear and detailed prompt, refining it repeatedly until it felt like a strong, business-focused design brief.
This same prompt was then used across all three platforms:
- Figma Make (primary focus)
- Magic Patterns
- Google Stitch
The idea was to write once, test everywhere, and see where each tool excelled – and where it fell short.
Magic Patterns: strong start, but limited
Magic Patterns immediately impressed me with its clean, responsive designs and solid UX structure. The results were interactive, modern, and could be exported directly into Figma for further refinement.
The main limitation was its credit-based usage model. As a free user, I quickly ran out of credits, which stopped me from iterating on the design or applying deeper levels of branding. There is a subscription available that offers more credits, but I did not explore that route for this test. It would be useful to understand the full cost of using the tool at scale, especially if aiming for production-ready outputs.
In short, Magic Patterns is great for quickly generating early concepts, but the access limits mean it may not be practical for creating a full, polished prototype without investing in a paid plan.
Google Stitch: great for flows, weak on visuals
Google Stitch excelled at mapping user journeys and structuring step-by-step logic. It worked well for early planning and wireframing, especially when visual polish was not yet a priority.
However, its UI output was basic. Buttons were static, fields lacked functionality, and customisation options were minimal. It simply was not designed to deliver the level of fidelity needed for a high-quality prototype.
Figma Make: From prompt to working prototype
Getting started
Using the refined prompt, Figma Make quickly generated a layout that covered all the key steps: login, brief upload, expert search, and final contact screens.
I asked it to add Google and LinkedIn login options for easier testing. These small iterations were surprisingly smooth, with the AI adapting the flow in minutes.
Uploading the brief
Users could upload a document or type their job brief. I added a “Just testing” shortcut, which opened a sample brief modal, making repeated iterations far easier.
Search and selection
The AI-generated list of experts was fully interactive. Profiles could be viewed, shortlisted, and moved through to an auto-generated email screen where content could be reviewed and edited before sending.
Responsiveness
Initially, the design only worked on desktop. With a single prompt, Figma Make updated the code to make every page responsive, which was impressive.
Iterating and branding
Branding proved to be its weakest area. Colour changes or style updates were inconsistent and required repeated prompting. Despite this, I was able to produce a fully styled, functional prototype with a few hours of extra effort:
- Functional prototype: ~1 hour
- Feature refinement: ~2 hours
- Styling and branding: ~4+ hours
Key learnings
- Prompt quality is everything – well-crafted instructions result in better outputs.
- AI can reduce design time – going from idea to prototype in a single day is achievable.
- Magic Patterns and Stitch have niche value – but are currently limited by access or fidelity.
- Branding remains AI’s weak spot – maintaining consistency is still challenging.
- Responsiveness is surprisingly easy – one good prompt made the Figma prototype work across devices.
The biggest takeaway? AI lets us go beyond simply describing an idea. We can now interact with a working version of it in hours rather than days.
Final thoughts
This experiment showed me that AI has become genuinely useful for producing real design outputs, not just concept sketches. With Figma Make, I was able to create an interactive, functional prototype without touching code or manually building layouts from scratch.
Some tools still lack polish, but the ability to bring a full product idea to life in a single day feels like a major shift in how designers can work. Instead of wondering “What if we tried this?”, I can now show it – instantly.