When a Jane Street developer wrote that he designs more with Claude than with Figma, the post collected 225 upvotes on Hacker News within 48 hours. There were 209 comments. Jane Street is one of the world's largest quantitative trading firms. The story of one of its developers building screens inside an AI chat window instead of Figma read as more than a convenience hack — it signaled that a genuine shift in professional tooling is already underway. Many of the people who responded were in exactly the same position: responsible for UI, working alone.
Building Screens With Prompts
His workflow goes like this. He describes the screen he wants in plain language — layout, colors, interaction behavior, component structure — and Claude generates HTML, CSS, and React code. He checks the result immediately in a browser, and if something's off, he asks for a revision. The cycle of framing in Figma, placing components, and producing a developer handoff file was replaced by a loop of text input and code review.
The biggest difference he noticed was the cost of switching contexts. Working in Figma, he moved back and forth between design mode and developer mode. Translating a UI mockup into code always introduced an unintended gap between intent and implementation. With Claude, that gap doesn't exist from the start. Describing the screen and generating the code happen simultaneously. He described the experience as "the feedback loop changed."
Others in similar positions chimed in across the Hacker News thread. Backend and full-stack developers who "don't have a designer but still need to ship a UI" said they recognized the feeling immediately. The recurring sentiment was that asking an AI precisely for what you need is faster than spending time learning Figma. The method proved especially effective for internal tools and admin panels — cases where functional completeness matters more than visual polish.
When to Question Whether Figma Is Even the Right Tool
There's a recurring theme in Korean solo-founder communities. Figma subscriptions get paid, but the features that actually see use amount to wireframing and rough mockups. The Figma Professional plan is $15 a month. Paying for it while barely touching the collaboration or prototyping features is a pattern that repeats across productivity tools.
One of the main reasons solo founders and solo PMs use Figma is to build a draft before handing off to a designer. But if there's no freelance or in-house designer in the picture, the Figma mockup isn't a handoff file — it's the final deliverable you also have to build yourself. In that situation, it's worth asking whether Figma is still the right choice.
There's a pattern that repeats whenever technical tools become broadly accessible. The tool itself doesn't get simpler — the interface for reaching the tool changes. It happened with aerial photography, data visualization, and 3D printing. Drone control software that once required a licensed operator became something filmmakers and farmers could handle once the interface changed. AI coding assistants are moving in the same direction. That doesn't mean they'll fully displace existing tools.
Why This Approach Doesn't Work for Everyone
Skepticism about this trend is real, and it showed up clearly in the Hacker News comments. Many replies offered specific counterexamples. The most common critique: "Claude-built UIs are functional but inconsistent." Button spacing varies subtly between screens, color tones don't match across pages, responsive behavior breaks on certain viewports. Building a single screen is fast; pulling multiple screens into a coherent system takes significant manual cleanup.
UX designers pointed to a more structural issue. Conversation with Claude is effective for modifying the current screen, but it isn't well-suited for designing a full user journey — stepping back to see the entire flow at once. It responds quickly to "change this button color," but shows real limits when asked to "structurally fix why users are dropping off at this step." Editing a screen and designing an experience are two different levels of work.
There's also the matter of who the Jane Street developer actually is. The reason this approach worked for him is that he can write UI code himself. Using Claude's output in production requires the ability to review generated code, spot problems, and direct revisions. A product manager or marketer with no development background attempting the same workflow won't necessarily get the same results. Writing a good prompt and judging the quality of what comes back are different skills.
What You Can Try Right Now
That said, there are practical takeaways from this experiment. Rather than framing Figma and Claude as competitors, the useful move is redistributing which tool gets used at which stage.
Early in a project — when the screen structure isn't settled yet — Claude can be faster. A prompt like "mobile order-confirmation screen, includes product name, quantity, total, CTA button, white background, indigo accent" produces a usable draft faster than staring at a blank Figma canvas for ten minutes. That draft can serve as the basis for team feedback, or as the point where you hand off to Figma for refinement.
On the other hand, if a design system is already in place, or if you're working on a product where cross-screen consistency is critical, it's better to maintain Figma's established style guide than to drop in Claude-generated components as-is. Claude re-evaluates style with every request. Over the course of a long project, that inconsistency accumulates.
It's also worth running the numbers. Figma Professional is $15 a month; Claude Pro is $20. If you're at an early startup stage where paying for both simultaneously isn't obviously justified, pausing Figma and using Claude to rough out your initial screens — then switching once the product becomes more concrete — is a legitimate option worth considering.
Count how many Figma files you made last month that actually turned into shipped features. Then ask whether the time you spent on them went toward sharpening ideas or toward operating the tool. I'd argue that question needs to come before the tool decision does.
