Build a scenario from workflows you already have. Chain the steps, then decide how much autonomy to give — pause for approval, pause to edit, or full auto.
I'm Yun, a strategic problem solver with an AI-augmented workflow, building design-minded products, and focusing on humanAI interaction.
Reimagining designers' workflow
Claude Code, CursorResearch partner · Concept pressure-test
Design work repeats in cycles: collect feedback, create tickets, hand off, repeat. Yet each AI session resets context. We re-explain, re-initiate, and manage the tool instead of working with it. I used Craft Agents firsthand and kept repeating manual prompts each session. Claude helped me pressure-test the concept and prototype interaction logic, but the core framing had to come from me. AI can generate solutions, but it cannot decide which problem is worth solving. Why does every session feel like the first one? The missing layer is not more capability. It is legible agent behavior within real workflows.
When you run a scenario, you see summarised steps. Dig into the reasoning if you need to. The agent asks clarifying questions before it starts, so it's not just autopiloting.
Sometimes a flow can't finish on its own. When that happens, it drops an action item for you, something to review, and confirm.
Want to try it yourself? Open the prototype. A demo with placeholder data only—not connected to production or real accounts.
Vertical Text - Figma Plugin
Cursor, FigmaImplementation · UX scoping
A Figma plugin that converts horizontal type into vertical layouts for East Asian typography — scoped, built, and shipped in under 48 hours. I used Cursor to work through the Figma plugin API, which I hadn't touched before. The constraint forced a clear division: AI handled the implementation, I stayed focused on the design problem, design details and edge cases. When it went wrong, it was always because my brief was vague, or chunky. The friction taught me more about how to collaborate with AI than the parts that worked smoothly. Listed on Figma Community. 120+ installs in the first three weeks.
Demo recording of Vertical Text in Figma.
Envisioning the future of Moodboard
CursorConcept prototyping
Moodboarding is fundamentally a pattern recognition task. Collecting and synthesising ideas is time-consuming, and the connections between them usually only exist in your head, not on the canvas. I built a small graph-based lab to test one question: if AI surfaces relationships between nodes as you add them, does it help you spot patterns faster while you're still the one making the decisions? Early stage by design — I wanted to validate the core behaviour before adding scope.
Showcasing how AI can sit in a designer's workflow in a visual way.
Bakevert: 0-to-1 Organic Growth
Cursor, Figma MakeImplementation
I used Cursor to handle scaffolding and repetitive logic, staying focused on product decisions — what to build, for whom, and why it would get found. The build took two weeks. Within three months: 42.6K organic impressions, average position 5.6 in Google Search. The growth wasn't engineered, it came from solving a real gap for a specific community. AI accelerated the execution; the product-market fit signal came from users.
Playful Experiments
Claude Code
Everyone says Claude Code has better aesthetics than other models. These experiments are my way of testing that, and honestly, enjoying the process of finding its edges.
Earth After Dark — NASA's night lights meet live flight paths, built by a planet enthusiast.
Blur in Motion — SVG motion experiments from Figma to code.
Motion study — radial burst and particle behaviour.
Motion.lib — AI-taught, human-curated motion for the rest of us.
Design Language Library — Understanding how design evolves from a history lover :)