Accelerating design workflows with LLMs: a PDP prototyping case study
In 2026, working smarter means using AI tools to move faster without sacrificing quality — if you know when and how to use them.
Major innovation is happening across the board in design technology, especially with tooling built around large language models (LLMs). We’re finally moving past the old roadblocks, like getting bogged down by manual production work or feeling limited by coding skills. These new LLM-based applications let designers move more quickly from ideation to validation, with fewer bottlenecks.
The number of dedicated deployment tools that are already available — along with Figma’s native LLM integrations — definitely signal a permanent shift, particularly toward AI-assisted prototyping and MVP scaffolding. To find out exactly how new tooling might enable quicker, more interactive prototypes, I put it to the test as part of an improvement initiative on our product detail pages (PDPs).
Solving the PDP video gap
When shopping online, users need to see a product in motion and from multiple angles to truly assess it and make a confident purchasing decision. To test this hypothesis, we recently analyzed data from our PDP and found a clear solution: autoplaying a video as the primary asset in the image gallery significantly drives both conversion and customer engagement.
The challenge here is one of scale: there’s a substantial content gap across PDPs — a shortage of rich, video assets where customers make buying decisions. Many PDPs still feature static images only. To optimise the experience for these pages, we needed to rapidly design and validate high-impact interim solutions while video production ramps up. This challenge presented a perfect testing ground for using AI-enabled tooling to discover which approaches were the most feasible and impactful.
Three hybrid approaches to test
The goal of this experiment was not to ship a completed design, but to help stakeholders visualise the different possible solutions and speed up the decision-making process. We had three concepts that we wanted to test and evaluate:
Auto-advance image gallery
This approach simulates video by automatically cycling through the existing gallery of static images. Dynamic transition effects also help the page to feel less static and more experiential.
Slideshow video
This option compiles existing static images into a pre-rendered video asset with polished transitions between images. The video is then placed as the primary PDP media file. Although it can’t display a garment in motion, it showcases the product without adding friction to the user experience.
AI-animated images
Because we have extremely high-quality baseline photography across our PDPs, the third concept involved using generative AI to animate static images. The technical challenge here is that a single source image doesn’t show the item in multiple dimensions; the tool needs to synthesise motion without hallucinating product characteristics or misrepresenting the item. Testing this was important for determining if this approach would be viable.
“Design-optimised LLM tools can help me to put my ideas into the hands of my audience — in a way that’s simpler, more polished and more interactive.”
The results: simpler for stakeholders and easy to iterate
To test and visualise these three concepts, I used LLM-based tooling to put together a functional prototype for each one. I started with the production Zalando app wireframes as a baseline. Then I used targeted prompts to define the motion mechanics, including the transition logic and speeds. I also added a configuration panel within the interface so that I and other users can toggle and test different transition styles in real time.
Get felixrossel’s stories in your inbox
Join Medium for free to get updates from this writer.
Although the objective wasn’t to deliver a finalised end-to-end user journey, my work with the LLM produced a relatively sophisticated prototype. Not only was I able to validate the three different concepts, this approach let me adjust the prototypes directly, without going back into my usual design tool and rewiring everything.
Using this new tooling also had other practical, operational benefits. During the review meeting, stakeholders could open the live prototype links directly on their phones and experience the concepts hands-on as I was presenting. One definitive takeaway for me from this exercise is that design-optimised LLM tools can help me to put my ideas into the hands of my audience — in a way that’s simpler, more polished and more interactive.
A new skillset for designers
Bridging the gap between concept and realisation has always been a core challenge of design. Traditionally, creating high-fidelity prototypes required heavy engineering support early in the process. Now, whether using native Figma LLM integrations or plugins for external environments, smart prompting lets us quickly validate our ideas independently, allowing us to bring stronger, more tangible concepts to our product and engineering partners. However, mastering this new workflow also requires adopting a few key guardrails:
- Data governance: Excellent data hygiene is an absolute must when using these tools: Avoid uploading proprietary assets, protect internal concepts from public exposure, and use enterprise-compliant tools.
- Code degradation: LLMs are more effective for prototyping when you use them to execute a defined vision rather than for ideation or general exploration. Frequent, radical pivots create technical debt and unstable code fragments, so start with strict requirements or finalized sketches.
This approach to prototyping is catching on more widely, as more and more designers begin to experiment and familiarise themselves with the technology. Ultimately, my experiments with LLMs for prototyping have had clear results so far: this is a powerful new technical capability that expands a designer’s autonomy and helps teams to accelerate the path to validation.
I set out to test whether AI-enabled tooling could help me speed up my prototyping process, and it certainly did. But not only that, it let me step into a more active role as an end-to-end creator and strategist, and share my insights with other designers. When we cultivate this new skillset in working with LLMs, it empowers us as designers to broaden our focus and have a deeper influence on product development at every step of the way.
How are you using smart prompting to speed up your workflows? Share your experiences in the comments!
