Case Study

Rebuilding a Desktop App

A mature desktop photo management application had grown over more than a decade. The features were powerful, but the interface carried years of accumulated patterns — and the underlying platform limited how far the experience could be taken. The decision was made to rebuild it from the ground up rather than continue patching it.

I own the UI/UX layer of that rebuild: designing the new interface, then writing and shipping the front-end code myself, with AI as a working partner rather than a code-completion tool.

This project is in active development and has not been publicly released, so the details here are intentionally general and the visuals are placeholders. I'm glad to walk through specifics in conversation.

Role:

AI Product Designer & Developer

Tools:

Figma, Claude AI, VS Code, GitHub

Timeline:

Nov 2025 – present

1

The Challenge

Rebuild a mature application without losing what made it valuable — and without shipping a downgrade to the people already using it.

Years of incremental additions had produced inconsistent patterns, competing conventions, and interactions that no longer matched how people actually work.
Existing users depend on specific workflows. A rebuild that quietly dropped capability would be experienced as a regression, no matter how much better it looked.
The interface had to be rebuilt on a new framework and hold up across desktop platforms, rather than being ported screen for screen.

2

Research & Discovery

Methods

Audited the existing application screen by screen, cataloguing every capability so nothing could be lost silently in the rebuild.
Traced the workflows people actually rely on, separating the paths in constant use from the ones that only looked important on a feature list.
Reviewed how comparable desktop applications handle the same problems, to avoid reinventing conventions users already know.

Key Insights

Parity is a design constraint, not a QA step. Deciding early what must carry over shapes every layout decision that follows.
Most of the accumulated complexity came from features being added beside each other rather than into a shared structure.
A rebuild is the rare chance to set conventions once, before a new decade of additions starts pulling against them.

3

Design Goals

Hold feature parity with the existing application, so the rebuild is an upgrade rather than a trade-off.
Establish one set of interaction and layout conventions, and apply them consistently across every surface.
Modernize the visual language without making the application feel unfamiliar to the people who use it daily.
Keep the design system close enough to the code that decisions can ship the same day they are made.

4

Designing & Building with AI

The part of this project that changed how I work.

I design the interface, then implement it myself — writing production front-end code with Claude and delivering it to the main release branch.
Closing the loop between design and implementation removes the handoff entirely. A layout question gets answered by building it, not by writing a spec about it.
AI is fast at producing plausible code, which makes design judgement more important, not less. Knowing what should exist is still the job.
Working this way means reviewing every change as a developer would — the output ships, so it has to hold up in review alongside the rest of the team's work.

5

Where It Stands

The rebuild is in active development. What is settled so far:

A consistent structure across the primary surfaces — navigation, panels, and detail views now share one set of conventions.
Interface work shipping continuously to the shared branch rather than accumulating in a separate design track.
Parity tracked deliberately as the rebuild progresses, so gaps surface while they are still cheap to close.

Impact & Reflection

Outcomes

Design decisions reach users as shipped code instead of stopping at a handoff document.
One consistent set of conventions replacing patterns that had drifted apart over years of incremental work.

Learnings

A rebuild is a design problem before it is an engineering one — what gets carried over is decided long before anything is written.
Designing and building the same thing changes the design. Constraints that would surface weeks later in handoff surface in minutes.
AI raised the value of taste. Producing something is now cheap; deciding what deserves to exist is the part that still takes a designer.