Detailed Analysis
Jane Street, the prominent quantitative trading and technology firm known for its rigorous engineering culture and heavy use of functional programming languages like OCaml, has published a blog post describing a significant shift in design workflow: a designer at the firm now relies on Claude — Anthropic's large language model, specifically in its Claude Code agentic form — more than Figma for design work. This represents a notable departure from established design practice, where Figma has been the dominant collaborative interface design tool for nearly a decade. The claim signals that AI-assisted code generation has matured to the point where it can functionally substitute for visual design tooling in at least some professional contexts.
The shift described reflects a broader pattern emerging across software organizations in 2025 and 2026: designers and engineers are discovering that generating HTML, CSS, and component code directly through large language models produces outputs fast enough and of sufficient fidelity to replace or significantly reduce the role of static mockups. Figma has traditionally served as a bridge between design intent and engineering implementation — a place where wireframes and prototypes are created before being handed off to developers. When an AI model can produce functional, inspectable, and iterable UI code directly from natural language prompts, that bridge becomes less necessary. Jane Street's internal tools and trading interfaces, which tend toward information density and precision over consumer-facing aesthetics, may be particularly well-suited to this workflow, as functional correctness often outweighs visual polish in that domain.
Jane Street's willingness to publish this observation carries particular weight given the firm's reputation for intellectual rigor and skepticism toward hype. The firm does not typically adopt tools for their novelty; its engineering blog posts tend to reflect genuine operational conclusions drawn from sustained use. The fact that a designer there has found Claude Code more useful than Figma in day-to-day work suggests the workflow has demonstrated concrete productivity advantages rather than merely theoretical appeal. This aligns with reports from other engineering-heavy organizations where the line between design and engineering is already thin — the AI collapses that distinction further by allowing design intent to be expressed directly as executable code.
More broadly, this development points to a structural challenge for design tooling companies like Figma. The value proposition of Figma has long rested on its role as a shared language between designers and developers — a place where ideas become concrete without requiring engineering involvement. Generative AI models like Claude undercut this by enabling either designers or engineers to produce concrete UI artifacts directly through conversation. While Figma has been adding AI features to its own platform, the competitive pressure from general-purpose LLMs capable of generating production-quality interface code is substantial. Jane Street's experience suggests that for teams with sufficient technical fluency, the friction of learning to design through AI prompting is already lower than the friction of maintaining a separate Figma-based design process.
The convergence of agentic AI coding tools with design workflows represents one of the more consequential near-term disruptions in software development practice. As models improve in their ability to interpret visual intent, maintain design systems, and produce accessible and responsive code, the traditional separation of design and engineering roles will likely continue to erode. Jane Street's public acknowledgment of this shift — coming from a firm not known for evangelical enthusiasm about emerging technology — adds credibility to the idea that this is not a fringe experiment but an emerging operational reality for technically sophisticated teams.
Read original article →