From mockups to working software: an AI-coded prototyping workflow
Role
Senior UX/UI Designer
Team
PM, designers, frontend and mobile engineering
Timeline
Mar 2026 – ongoing
Platform
Web, iPad, and mobile prototypes (Claude Code, Cursor, Vercel)
Scope
Team-wide change to how design is reviewed and handed off
TL;DR
The stake
Static mockups could not explain dense table and Gantt interactions. Handoffs needed long explanations, and problems surfaced late.
What I did
I helped move the team from static Figma mockups to AI-coded working prototypes, and built the habits around them: live links, async reviews, and design-system documentation the AI can read.
How I led
I built and maintained the prototypes the team reviews against, named the new risk they created, and mentor colleagues across the company on these AI workflows.
Results
Idea to testable prototype in under a day.
Feature shipping speed
About 3× faster, together with Design System 2.0.
Idea to testable prototype
from idea to a working prototype
live prototypes: web, iPad, and mobile
Shipped since September 2025
features designed and shipped across web, mobile, and iPad
async review rounds in three weeks for one view
Outcomes
What else changed.
Developers and stakeholders now validate real interactions before build.
A takt-planning prototype from this workflow was shown at an industry conference in October 2026 and led to follow-up conversations with prospects.
01 · Context & problem
Mockups couldn’t show behavior.
In early 2026 I was redesigning the Lookahead, a table beside a Gantt chart with inline editing, right-click actions, and hover controls. A static mockup shows what a screen looks like. It does not show how a dense interaction behaves.
In March 2026 I put the redesign into a coded prototype for handoff. There were many design differences, and a prototype explained them with context more efficiently than annotated screens.
02 · Constraints
Designers aren’t engineers, and AI invents.
Designers are not engineers. The workflow had to work for people who think in Figma.
Prototypes are not production. Engineering still builds the real thing in Storybook and the product codebase.
AI invents components. Without guidance, an AI tool creates a new button every time it needs one.
03 · My role & influence
Three prototypes, and the habits around them.
I worked with a PM who also prototypes in code. I built and maintained three prototypes, for the web Lookahead, the iPad app, and the mobile app, and I used them for handoff and design review. I mentor colleagues across the company on these multi-agent AI workflows.
04 · Strategy
Prototype the behavior, review it where it runs.
Prototype the behavior, then review it where it runs. Keep one live prototype per platform, branch it for each exploration, and give the AI the design system so new work starts from real components.
05 · Key decisions & trade-offs
Seven decisions behind the workflow.
1. Prototype the behavior, not the screen.
The first coded flow covered what a mockup could not: a new work-breakdown reference with a popover, action buttons that appear on task cells, and right-click menus in the table and the Gantt. Engineers saw the behavior before they built it.
2. One live prototype per platform.
Web, iPad, and mobile each have a prototype deployed to a permanent link. Each exploration is a branch with its own preview link. A switch in the iPad prototype shows a feature from either user role.
3. Give the AI the design system.
I merged the button documentation page into the Lookahead prototype. When anyone creates a new design, the AI references real components instead of inventing new ones. We are building this out for the rest of the basic UI components.
4. Use the AI to audit.
I asked the AI to go through the prototype and document every button variant in use. The audit surfaced several buttons that needed to be removed.
5. Review async, with a recording and a link.
Each design review post has a short screen recording and a live prototype link. For the daily view of the Lookahead, I ran five rounds in three weeks, including one with three versions side by side.
6. Write specs inside the prototype.
For table empty states, I built the spec as a page in the prototype, next to the working examples it describes.
7. Name the new risk.
Designers now refine components inside prototypes, and those changes do not always reach Figma or Storybook. I raised this with our engineering lead in June 2026 and proposed refining components in one place, then updating every prototype at once, like a pull request.
06 · Reflection
A second source of truth.
The prototypes made decisions concrete, and they created a second source of truth. Keeping Figma, the prototypes, and Storybook in sync is the open problem. A reliable path from prototype back to Storybook is what I would build next.