AI Flips the Economics of Design and Code

The old cost model of design work is gone. Historically, code was expensive—complex, laborious, and painful to revise—while design exploration felt cheap. Teams could sketch out dozens of static mockups quickly and converge on a direction over time. AI has inverted that balance. Now it is often faster to build a functional wireframe than a static one, shifting iteration from layout to actual behavior. Gui Seiz, design director of AI at Figma, describes how going from code to canvas helps teams avoid getting locked into a single visual direction.

For Alex Kern, a software engineer at Figma focused on AI-powered creation, the change is about more than speed. Keeping Figma in sync with a codebase was always tedious work: maintaining consistent structure and aligning decisions across disciplines. AI models now treat code as a creative medium, making the translation between canvas and code more semantic and far less mechanical.

Access, Not Philosophy

For years, the debate was whether designers should learn to code. Kern argues AI has made that question obsolete—the more pressing one is why a designer wouldn't ask AI to code for them. This matters most for access. A designer working outside a company's internal design system can now pull the live product into Figma as editable frames and start from there, without waiting for the right permissions or assets. The barrier to entry has dropped considerably.

That newfound access also reshapes how people learn. The steep learning curve that once separated designers from code has flattened into a ramp. Instead of abstract bootcamps, AI provides contextual instruction grounded in whatever a person is actually building. Seiz notes that concepts like routes and React finally make sense when learned as part of concrete work, rather than in isolation. Designers who once stopped at the edge of their technical knowledge are now experimenting with shaders, 3D tools, and custom utilities.

Working Across the Roundtrip

The most practical shift is the arrival of a genuine roundtrip workflow. In the past, Figma was where teams aligned; code was the source of truth, and the bridge between them was screenshots. A designer would capture an app, import it into Figma, mark it up, and pass the work back for manual reconstruction. Now, with Figma MCP, production code can move onto the canvas and back again as editable designs. Storybook components, blog posts, graphics, and real tables of live data can be imported quickly, allowing design and build to stay in sync with coverage close to 100%.

This overlap gives teams more flexibility about where they start. Designers can make copy and design changes directly in code, while developers can pull product states onto the canvas to check their work. Kern points out that code tends to be single-player, and AI models trained on an existing codebase often bias toward preserving the status quo. A designer working in Figma can break free of that, exploring a radically different direction without fighting the accumulated momentum of the current implementation.

Code Is No Longer a Bottleneck

Time was always the engineer's most binding constraint. With AI agents, that constraint loosens significantly. Engineers now take on projects that previously would have been an irrational use of time: rewriting entire open-source projects in new languages, building browsers from scratch, or clearing decade-old backlogs with a tiny team. Interestingly, the models themselves don't have a sense of time—Kern jokes that Claude Code might estimate three months for a task it will finish in five minutes. When time pressure fades, the design question shifts from "How long will this take?" to "What do we want to build?"

The New Differentiator

When everyone has equal access to the same AI tools, taste alone won't be enough to stand out. Seiz argues that curiosity becomes the real differentiator. AI functions as a patient tutor, handling the setup that once required deep familiarity with toolchains and syntax. Designers who keep asking questions—about shaders, about 3D, about their own products—will define the field ahead.

Kern sees the opportunity on the other side as well. For software to distinguish itself, it will need to lean on design that experiments with the form itself. Interfaces that blend engineering and design can deliver what only software can create: live data visualizations, real-time comprehension of intent, generative UI, and ephemeral tools. The underlying technology already exists. What remains is for creative people to decide what to do with it.