The setup
In February 2025, a small internal team was formed in my current workplace to explore an AI product focused on crypto and privacy. The team was lean: five engineers and me as the only designer, operating more like a startup squad inside a larger company.
I have designed digital products and experiences from 0→1 since 2017. With AI tools advancing quickly and product development moving faster, designer roles and workflows are constantly changing. These are some observations and learnings from designing during the current AI wave.
Requirements were often vague, if they existed at all.
Most features started as a sentence in Slack or a quick conversation between the team and stakeholders. No brief, no spec.
My loop became: quick Figma mockup → clickable prototype (Figma or v0/Claude) → screen recording → post in Slack. Sometimes within the hour. The goal wasn’t to get it right on the first try. It was to give the team something concrete to react to before writing production code.
Small improvements compound fast
During growth stages, there were endless opportunities for small UX improvements. The AI model selector redesign was one example of this:
The first version was a fairly standard dropdown. Functional, but limited once the number of models and media types started expanding.
Over time, the product outgrew the interface, the list of models kept growing, and the output type (text/image/video) was invisible.
I designed a new selector with simple and advanced modes: Autorouter for users who just want an answer, manual selection for those who want more control.
Design isn’t always the starting point
When we decided to show the exact credits required for prompts and image generation, one of the engineers added a rough credit estimate to the bottom-right corner of the chatbox.
Functionally it worked, but visually it felt disconnected, floating like an orphaned UI element.
I redesigned it based on his version. I moved it underneath the chatbox as a banner, gave it a soft background, and made it slide down subtly when the user starts typing.
Most of my growth-stage work looks like this. An engineer solves a problem functionally; I resolve it visually and make it legible. Treating these as collaborations rather than handoffs is the only way the loop stays fast.
Not everything ships
Some concepts never made it into production. But unused work still matters. In fast-moving teams, concept work often acts as future infrastructure. An idea that feels unnecessary today might suddenly become relevant months later once the product direction changes.
Closing
The biggest shift for me wasn’t the tools. It was the adaptability mindset.
Being a designer building products in the AI era, especially in startups, feels less about protecting a fixed process, and more about staying close enough to the build to keep changing it.
That flexibility has become part of the craft.