I rebuilt Windows 95 in a weekend. The hook isn't the nostalgia. It's that I treated it as a design system, not a project.
Windows 95 already shipped a complete design language. Buttons, dialogs, window chrome, type scale, color tokens, all designed, all documented, all coherent. Decades ago. By people who took design systems seriously before the term existed.
The community took that work and turned it into open-source code. React95 for the widgets. Webamp for the actual Winamp. A design system plus a working app, both sitting in public repositories.
The stack:
- React 19 + Vite
- React95 (core + icons) for the design system: windows, buttons, taskbar, cursors
- Styled-components for the Win95-specific 3D beveled borders
- Webamp embedded for actual music playback
- Vercel for deploy
The split between leverage and build was the design decision:
- Leveraged: every UI primitive, every cursor, every icon. React95 gives the entire design language as components. I touched zero pixels of chrome.
- Built: window management (drag, minimize, maximize, z-index, taskbar), an iframe-based Internet Explorer with address bar and CRT scanline effects, and a dial-up loading experience that opens 40+ archived 90s sites.
I split the AI work deliberately. Opus 4.5 got the architectural pass: one careful prompt, out came a project structure that didn't fight me later. Sonnet 4.5 handled iteration: describe behavior, get the implementation, test, refine. The speed wasn't AI-magic. It was the right model for the right shape of work.
The fastest way to ship something polished isn't to start from scratch, and isn't to assemble from parts either. It's to find the design system that already exists, leverage every primitive it gives you, and spend your engineering on the part that actually requires invention.
What 90s app would you rebuild?