The designer argues that Claude Code has become their primary canvas because the design artifact and the production artifact are the same React tree at Jane Street. Figma was always a translation step, and translation steps are where friction and bugs accumulate — editing TSX directly eliminates that lossy handoff entirely.
The editorial frames the post's real claim as a structural critique of the design-tool industry: fifteen years of products from Anima to Figma Dev Mode have tried to close the handoff gap and shipped slop because the gap itself is the problem. Claude Code wins not by being better at visual design but by collapsing the workflow into the medium where decisions actually ship.
The editorial emphasizes that Jane Street ships internal software to a small sophisticated user base with no marketing site, brand guidelines, or external design system to reconcile. That context — not the AI tooling — is what makes the workflow viable, suggesting the result may not generalize to consumer product teams with broader design surface area.
A designer at Jane Street — the OCaml-obsessed quant trading firm — published a post on the company's tech blog titled "I design with Claude Code more than Figma now." It's not a hot take from a Twitter influencer. It's a working designer at a firm whose entire engineering culture is built on the premise that the right tools compound. The post walked through real workflows: prototyping internal trader-facing tools, iterating on layouts by editing TSX directly, and using Claude Code as the surface where design decisions get made.
The headline number isn't a benchmark — it's the verb. "Design with." Not "generate from" or "hand off to." The designer is treating the LLM-driven code editor as the primary canvas, and Figma as a sketchpad they occasionally open to think visually.
The Jane Street context matters more than the AI-tooling angle: this is a firm that ships internal software to traders, where the design artifact and the production artifact are the same React tree. There is no marketing site, no brand guideline, no design system that needs to be reconciled with a separate engineering implementation. The designer's job is to ship usable interfaces to a small, sophisticated user base. Figma was always a translation step in that workflow, and translation steps are where bugs and friction live.
The interesting claim in the post isn't that Claude Code is better at visual design than Figma. It clearly isn't. The claim is that the visual-design step was a smaller fraction of the actual job than the design-tool industry has spent fifteen years pretending.
Design handoff has been the lossiest step in product development for a decade — every team has felt the gap between the Figma file and the shipped UI, and every "design-to-code" tool that tried to close it has shipped slop. Anima, Locofy, Builder.io, Figma's own Dev Mode: each tried to make the handoff cleaner. None of them got rid of the fundamental problem, which is that a designer expressing intent through a vector tool and an engineer expressing implementation through a component library will always disagree at the seams.
Claude Code's approach is structurally different. The designer writes prompts, reads the diff, runs the dev server, and iterates. The output is the implementation. There is no handoff because there is no second artifact. Removing the handoff doesn't make the AI a better designer — it makes the designer's intent survive the trip to production. That's a meaningful shift even if you think LLM-generated layouts are mediocre.
The community response on Hacker News skewed predictable: half the thread argued this only works because Jane Street designers can code, half argued that all designers should be able to code, and a smaller faction pointed out that the post is suspiciously well-aligned with Anthropic's go-to-market story. All three are partially right. The post is unusually clean evidence for a thesis Anthropic has been pushing — that Claude Code is a general-purpose creative tool, not just a coding assistant. But Jane Street isn't running Anthropic's marketing for them. The firm has a 20-year track record of writing detailed engineering posts about their actual workflows, and they don't ship puff pieces.
The more honest read: this is a strong signal for one specific quadrant of design work — internal tools, dashboards, admin interfaces, anything where the user is sophisticated and the visual surface is a thin layer over data — and a weak signal for everything else. A designer shipping a marketing landing page is not switching to Claude Code next quarter. A designer shipping a trader's order entry screen at a quant firm clearly already has.
If you run a team that ships internal tools — admin panels, ops dashboards, B2B SaaS configuration screens — the Jane Street workflow is worth a one-week experiment. The test isn't whether the AI designs well; it's whether your designers ship faster when they skip the Figma-to-React translation step. Pick one project, have the designer work in Cursor or Claude Code with a checked-out copy of your component library, and measure cycle time from problem statement to merged PR.
The harder question is what this does to design-system investment. Component libraries get a lot more valuable in a world where the designer is editing component instances directly. The system stops being a reference document a developer translates from and starts being the primitive vocabulary the designer types in. Teams that have under-invested in design systems will find their AI-driven design output looks worse than their Figma output. Teams with strong systems will find the opposite.
For design hiring, the implication is uncomfortable but obvious: the floor on technical literacy for designers working on internal software is going up, not down. "Designer who can code" was a unicorn job description in 2018; in 2026 it's the description of every designer at a shop like Jane Street, and that's the cohort that just got a force multiplier. Designers who can't ship a diff will increasingly be confined to surfaces where pixel-perfect visual craft still dominates — brand, marketing, consumer onboarding.
Figma isn't going anywhere. The company just IPO'd at a valuation that assumes it's the default surface for design work, and the consumer-app and brand-design quadrants are not under serious threat from a code editor. But the most technically rigorous design teams in software are the leading indicator, and one of them just published evidence that the design tool they reach for first is a terminal. Watch for the next two or three posts like this from places that aren't Jane Street. If they show up, the design-tool category is in the middle of a quiet bifurcation: code-first tools for software that ships as code, vector-first tools for everything else. Figma's bet has to be that it can be both.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.