From Research to Prototype in 2 Days with AI
Figma MCP
Claude
Lovable
Web3
A 0-to-1 case study on a Web3 consumer app rich in technology but missing a market. Using Claude and Figma MCP for discovery and sense-making, then Lovable for live prototypes, I went from "where does this even fit?" to testable product directions in a fraction of the usual time. The point isn't the speed, it's that AI let me run strategy, design, and prototyping as one continuous loop.
Names and certain details are masked due to an NDA.
Introduction
This case study covers how I ran the full 0-to-1 loop as one continuous, AI-native process. Discovery that would normally take weeks took a day
Role
Lead product designer, end to end. I owned market discovery, product strategy and positioning, design direction, and prototyping, using Claude for research and sense-making, Figma MCP for organizing and exploring directions, and Lovable for live, testable prototypes.
Challenges
Project had many innovative technology but lacked a clear market position. Before designing any UI, the key question was: where does this product belong, and why would users need it? Solving that required market research, user-case mapping, and competitive analysis to identify a strong, defensible niche.
No market position
The product had strong technology but no defensible niche, no clear answer to where does this belong and why would anyone need it here?
Strategy before UI
The first problem wasn't a screen or a flow; it was positioning. Any design work risked being wasted until the market gap was found.
A crowded, fast-moving market
The crypto space is saturated. Entering the wrong segment meant direct competition with no advantage.
Compressed timeline
Weeks of market research, user-case mapping, and competitive analysis had to happen fast, without thinning the rigor.
From insight to something testable
A market thesis is only worth as much as the evidence behind it — it needed to become a working prototype people could react to, not a slide.
Discovery in a day / Claude
I ran the entire discovery layer through AI in a single day: market research, user cases, and competitive landscape analysis. The goal was specific — not "understand the market," but find the gap: a position where competition was low, the need was real, and entry would be a strong play.
The output was a concrete finding — a hole in the crypto market where moving in would be a genuinely strong move for Project Nova.
Judgment: What mattered wasn't that AI made research faster. It's that compressing discovery to a day let me treat strategy as something I could iterate on — running, testing, and discarding market hypotheses the way I'd normally only iterate on UI.
Project screens hidden under NDA
Sense-making and variants / Figma MCP + Claude
With Claude and Figma MCP I organized the research into something I could reason about visually, and generated several directional variants. Seeing the options side by side, not as abstract notes but as structured, comparable directions, let me reason through the whole picture and pressure-test where the product should go.
Judgment: The variants were there to think with, not to ship. AI's real gain here was breadth: more directions made visible and debated before committing to one.
Project screens hidden under NDA
From research to a build-ready spec / Claude
Once a direction held up, I used Claude to write a complete Lovable prompt, one that carried every component the research had pointed to, so the build reflected the strategy rather than drifting from it.
Judgment: This is the step most "AI prototyping" stories skip. The prompt is the design spec. Getting it right, encoding the research-backed decisions into it, is where the designer's judgment lives, not in the tool that renders it.
Placeholder, real screens under NDA
Live, testable prototypes / Lovable
Once a direction held up, I used Claude to write a complete Lovable prompt, one that carried every component the research had pointed to, so the build reflected the strategy rather than drifting from it.
Judgment: This is the step most "AI prototyping" stories skip. The prompt is the design spec. Getting it right, encoding the research-backed decisions into it, is where the designer's judgment lives, not in the tool that renders it.
One real friction
Where did the AI do something you didn't intend?
e.g. Lovable assuming a flow or adding a component the prompt didn't specify, or a research conclusion that looked confident but needed a human gut-check. Name one concrete moment and how you caught it.
Judgment: This is where AI's fidelity to intent breaks down, it interprets, it doesn't just execute. Catching that gap is the designer's job, and naming it honestly is what makes the rest of this story credible. (A clean success story reads as marketing; one real friction reads as experience.)
Where It's Headed?
AI doesn't replace the designer; it dissolves the walls between phases. Market discovery, strategy, design, and code-backed prototyping stop being separate handoffs and become one loop a single person can run.
The new skill isn't knowing the tools, it's judgment: which question to point AI at, when a direction is worth building versus discarding, how to encode strategy into a prompt so the build doesn't drift, and where not to trust the output. The designer becomes the person who runs the loop and owns the decisions.