Google's Stitch 2.0 promises to turn finished designs into working code, but the real story is more nuanced than "AI replaces developers."
Design handoff has always been a point of friction. Files move from Figma to a developer's editor, and somewhere in that transition, spacing gets approximated, hover states get reinvented, and intent gets lost. Stitch 2.0 is Google's attempt to close that gap. Here's what it actually does, where it helps, and where designers still need to stay hands-on.
What Google Stitch 2.0 Actually Does
Stitch 2.0 is a Google machine learning model built to convert visual designs, typically pulled from tools like Figma, into functional, web standards-compliant code. It accepts both raster and vector inputs and outputs HTML, CSS, and JavaScript.
What sets it apart from earlier code-generation attempts is its focus on preserving design intent. Rather than spitting out generic markup, it aims to retain spacing, typography, and component hierarchy, the details that usually get flattened in manual translation.
The Design-to-Code Problem It Attempts to Solve
Traditional handoff forces developers to interpret static files and make implementation calls the designer never explicitly made. Design systems help standardize this, but tokens still require manual translation into code.
The bigger issue is context loss. Hover states, responsive breakpoints, and interaction logic often live only in a designer's head or a Slack thread, not the file itself. Teams without dedicated design engineers absorb this as inconsistent quality and repeated rework.
Current Limitations and Realistic Expectations
Stitch 2.0 performs well on standard patterns: buttons, cards, grids, forms. It struggles with complex interactions, custom logic, and anything requiring real judgment.
Generated code needs developer review before it ships. Static designs can't tell the model how to handle state management, API integration, or dynamic content. And accessibility, ARIA labels, semantic HTML, keyboard navigation, depends entirely on the quality of the source file and the model's training, not automatic inference.
How It Fits Into Modern Workflows
The strongest use case is accelerating prototyping and early iteration, not replacing developer judgment. Stitch 2.0 generates component instances; it doesn't build the design system itself.
It pairs naturally with tools like Figma's Dev Mode that are already becoming standard in mature teams. Used well, it reduces time spent on boilerplate so developers can focus on logic, performance, and the edge cases that actually require expertise.
When to Use It, and When Not To
Good fits: marketing sites, landing pages, brochure-style sites, admin dashboards with standard CRUD patterns, and design system component exports.
Poor fits: applications with complex state, real-time collaboration, custom animation, or strict accessibility requirements.
Output quality tracks directly with design file discipline. Teams using structured components and consistent naming get usable code. Messy, ad-hoc files get messy, ad-hoc output.
The Broader AI-to-Code Trend
Stitch 2.0 enters a category that already includes Figma Copilot, Locofy, and Builder.io, each with different strengths and framework preferences. Quality varies widely across tools, and most excel in a narrow niche rather than across the board.
Adoption pays off for teams with disciplined design systems already in place. Treating any of these tools as a shortcut for poor communication tends to produce disappointing results.
What This Means for Designers and Developers
AI code generation is a productivity layer, not a substitute for design thinking or engineering skill. As routine translation gets automated, the real value of design work shifts upstream toward strategy and systems thinking.
Healthy teams will treat Stitch 2.0's output as a conversation starter, "here's the intent, let's refine it," rather than a final deliverable. Track time saved, rework cycles, and design system compliance over time rather than judging the tool on raw speed alone.
The takeaway for design leads: audit your current handoff process before adopting any AI code tool. If your design system and naming conventions aren't solid, fix that first. Stitch 2.0 amplifies good process; it doesn't create one.
