Figma built its category on a simple assumption: design and code are separate disciplines. Designers create the blueprint; developers build it. A designer at Jane Street - the quantitative trading firm known for unusually rigorous engineering culture - published a post this week making the case that this pipeline is collapsing for a specific type of practitioner. They now reach for Claude Code before opening Figma.
The argument isn't that AI tools are better at visual design. It's that for people who design and build their own products, the Figma step adds a translation layer that didn't need to exist. You create an artifact - the mockup - that immediately has to become a different artifact: the code.
What the Code-First Design Loop Looks Like
Claude Code writes actual HTML, CSS, and component code - not screenshots or design files. The workflow: describe a layout or component in natural language, get a working implementation, open it in a browser, see what needs adjusting, describe the change, repeat. There's no "hand off to yourself" step.
For someone with enough coding fluency to read and modify generated code, this loop can be genuinely faster than visual prototyping. Especially for components you'd describe verbally anyway - "a sidebar with collapsible sections and a sticky header" renders in roughly the time it takes to drag elements into Figma.
This isn't specific to Jane Street. Developers have been using Claude Code, Cursor, and Bolt.new to generate UI components for over a year. What's shifted is the confidence level - practitioners are committing to this workflow for entire products, not just one-off experiments. A post titled "I design with Claude Code more than Figma now" implies the workflow is reliable enough for real work, not just prototyping.
One honest limitation: Claude Code handles structure and layout logic well, but visual refinement is less consistent. Tight spacing, custom animations, brand typography - those still require manual CSS and design taste. You're generating iterable starting points, not polished interfaces.
Where Figma Isn't Going Away
Figma's core use cases don't disappear because solo practitioners prefer code-first iteration. Collaborative review from non-technical stakeholders, design system governance, handoff specs for a separate dev team, clickable prototypes for client approval - these all require a shared visual artifact that people without coding knowledge can interact with and annotate. Generated code doesn't serve that purpose.
Figma's enterprise business - large teams with distinct design and engineering functions - is probably fine. Its grip on solo practitioners and small technical teams that don't hand designs to anyone else is where the pressure is building.