I design with Claude more than Figma now
AI-assisted “vibe coding” tools like Claude Design are starting to replace Figma and traditional wireframing for many developers and product teams, who find they can generate interactive front-end prototypes far faster by prompting an LLM than by hand-drawing in design software. Commenters describe real gains in speed, iteration, and empowerment for non-engineers, but also warn about generic, samey UIs, poor handling of edge cases, and the risk that stakeholders mistake brittle prototypes for production-ready systems. A recurring concern is that outsourcing early design and architectural thinking to models can erode clarity, increase maintenance burdens for engineers, and deepen tensions between rapid experimentation and responsible software quality.
Use of Claude/LLMs for UI Design & Prototyping
- Many commenters now use Claude Design / Claude Code (and similar tools) to generate frontends, wireframes, and working prototypes, often before or instead of Figma.
- Common pattern: prototype in code with an LLM, then refine visually in Figma or an existing design system.
- Some team workflows: designers/product generate interactive prototypes with LLMs, engineers later re‑implement “for real,” using prototypes as living proposal docs.
Prompting, Creativity, and Generic Outputs
- Without detailed aesthetic direction, models default to safe, trope-heavy designs (Tailwind/Bootstrap clones, “boarding pass” travel UIs, etc.).
- Users report more originality when they:
- Provide unconventional reference sites or UI examples.
- Specify fonts, layout constraints, and brand/style cues.
- Even then, outputs often feel visually similar and struggle with layout boundaries (e.g., overlong slides).
Impact on Workflow, Roles, and Specs
- Some designers/PMs feel empowered to “vibe code” UI and logic, reducing reliance on engineers for early iterations.
- Frontend engineers report new burdens:
- Reviewing messy AI-generated PRs.
- Distinguishing intended behavior from accidental “slop.”
- Dealing with fewer written specs and more reverse‑engineering of prototypes.
- Others like that prototypes make requirements more concrete than verbal/text specs alone.
Quality, Maintainability, and Risk
- Several engineers compare LLM output to a junior dev: useful, but full of hidden bugs, missing edge cases, and wrong assumptions.
- Pushback against treating LLMs as a “higher abstraction layer” like compilers, since output isn’t deterministic or reliably correct.
- Concern that non‑technical stakeholders will think prototype code is “95% done” and demand quick productionization, underestimating robustness, security, and data‑integrity work.
Broader Attitudes Toward AI & Tools
- Split sentiment:
- Enthusiasm about faster iteration, infinite patience for tweaks, and enabling small orgs or NGOs to replace overpriced contractors.
- Skepticism about hype, “cult-like” inevitability narratives, and corporate marketing (including from trading firms invested in AI).
- Some see AI as particularly well-suited to disposable UI, images, and low‑risk web pages; others worry it erodes deep thinking and design rigor.
- Discussion touches on Figma’s future, perceived weak AI features there, and early signs of pressure on traditional design and website-building tools.