Roundups
Figma AI vs. Dedicated AI Design Tools: How They Actually Compare
Figma disabled its "Make Designs" AI tool in 2024 after it produced app designs nearly identical to Apple's Weather app, with CEO Dylan Field publicly saying "it is my fault for not insisting on a better QA process." Practitioner feedback since has been sharper still: one commenter called Figma Make "incredibly underwhelming... basically just Gemini 3 Pro with a design prompt," criticizing it for generating pages that ignore existing components and design tokens. Dedicated tools like Claude Design differentiate specifically by importing an existing design system on onboarding and applying it automatically to every future generation — though the emerging 2026 consensus among practitioners is that most product teams end up using both, for different stages of the same workflow, rather than picking one.
Last updated Aug 21 · 19 min read
The incident: Figma's AI copied Apple's Weather app
The most consequential single event in this space is a real, documented product failure, not a hypothetical risk. Figma disabled its "Make Designs" AI tool after it produced app designs that were nearly identical to Apple's Weather app (404 Media).
Figma CEO Dylan Field responded directly and publicly, saying: "Ultimately it is my fault for not insisting on a better QA process." The feature was paused pending further quality-assurance work rather than being quietly patched and left running (404 Media).
This matters beyond the specific incident because it directly informs the practitioner skepticism discussed below — a tool from the company that essentially defines the modern design-tool category shipped an AI feature with a QA gap severe enough to require disabling it publicly. That's the backdrop against which every subsequent comparison to dedicated AI design tools has been made.
What Figma Make actually does
Setting the incident aside, Figma's own positioning of Make is straightforward: prompt-to-prototype generation that stays visually editable directly on the Figma canvas, and can pull in "Make kits" or existing Figma frames as design-system context for generation (Figma).
The intended value proposition is clear: designers who already live inside Figma files get AI generation without leaving their existing workflow, and can refine the output with the same point-and-click tools they already know — adjusting spacing, typography, hierarchy, and alignment without switching contexts.
The core practitioner critique: design-system fidelity
The most consistent, most specific criticism of Figma Make in practitioner discussion is that it doesn't reliably respect an existing team's actual component library and design tokens. Direct quotes from Hacker News discussion capture this precisely: Figma Make "generates pages that ignore your components, skip your tokens," treating "every screen like a fresh snowflake" instead of reusing what the team has already built (HN discussion).
One commenter went further on the underlying model quality itself, describing the tool as "incredibly underwhelming... basically just Gemini 3 Pro with a design prompt" — a characterization that, if accurate, suggests the gap isn't really about Figma's product wrapper at all, but about the underlying model powering the generation (HN discussion).
Why this specific criticism matters more than a generic "the output looks bad" complaint: for any team with an established design system, a tool that ignores existing components isn't just producing lower-quality output — it's actively creating rework, because someone now has to manually reconcile AI-generated screens with the team's actual token and component standards after the fact. A tool that respects the existing system produces output that's immediately usable; one that doesn't produces output that needs translation work before it's usable at all.
Claude Design's specific differentiation
A direct comparison published by a design-focused blogger argues that Figma Make runs on a weaker underlying model than Claude Design, and that Claude Design's specific mechanism — a one-click import of an existing design system during onboarding — produces more professional, on-brand results as a direct consequence (Martin Alderson).
The mechanism, per this account and corroborated by 2026 comparison coverage: during onboarding, Claude Design scans a team's codebase and design files to learn their brand tokens, components, and voice, then applies that learned system to every subsequent generation (Eigent.ai; Magic Patterns).
The same source argues this specifically enables non-designers to produce professional-looking output that matches existing corporate design standards — a meaningfully different value proposition than "generate something novel and pretty," which is closer to what a fresh-canvas prompt-to-prototype tool without system-import does (Martin Alderson).
Native handoff to code. Claude Design's workflow also includes a native handoff to Claude Code, letting an engineering team pick up and continue the generated work directly — a workflow connection that ties design generation more tightly to implementation than a Figma-canvas-only workflow does (Eigent.ai).
The economics: why smaller competitors have an edge
Beyond the feature-level comparison, there's a structural economic argument worth understanding. Figma pays for the same underlying AI inference costs it's competing against — every Figma Make generation costs Figma real compute — while smaller, more focused dedicated competitors can operate with near-zero marginal cost per shipped feature improvement, since they're often building thinner, more targeted products on top of the same class of underlying models (Martin Alderson).
This is presented as commentary/analysis rather than verified financial data from either company's actual unit economics — treat it as a plausible structural explanation for why dedicated AI design tools can iterate quickly and cheaply, not as an audited comparison of Figma's and any competitor's real costs.
Practical examples
Real, sourced example — the Apple Weather app copy incident. This is the clearest, most concrete evidence in this entire topic area: a real AI design tool from a major company produced output essentially indistinguishable from a specific, well-known existing app, publicly reported, publicly acknowledged by the CEO, and resulting in the feature being disabled (404 Media). Any evaluation of an AI design tool's originality and QA rigor should treat this as the concrete bar to check a candidate tool against, not an abstract concern.
Real, sourced example — the shared-defaults problem. One commenter attributed the sameness of a lot of AI-generated UIs — regardless of which specific AI design tool produced them — to many teams and tools defaulting to the same off-the-shelf CSS/component library (shadcn was named specifically) as a shared starting point (HN discussion, cited in research brief). This is a useful, separate explanation from the design-system-fidelity critique above: even a tool that doesn't ignore your components can still produce generic-looking output if its underlying training or default templates lean heavily on the same popular open-source component libraries everyone else's AI-generated UI also leans on.
Illustrative example (hypothetical, clearly labeled) — a startup without a dedicated designer. A small SaaS startup with no in-house designer needs a professional-looking marketing site and app UI quickly. Per the practitioner accounts cited above, a tool like Claude Design — which imports an existing design system (or, absent one, applies a coherent one automatically) and is specifically credited with letting non-designers produce corporate-standard output — would plausibly outperform a prompt-to-prototype tool that generates each screen from scratch with no system awareness. This is an illustrative application of the sourced comparisons, not a documented specific case.
Data and evidence
– Figma disabled its "Make Designs" AI tool after it produced designs nearly identical to Apple's Weather app; CEO Dylan Field publicly attributed this to insufficient QA process (404 Media).
– A named alternative, Magic Patterns (YC W23), drew 185 points and 114 comments in Hacker News discussion — a substantial engagement level indicating genuine practitioner interest in AI design tool alternatives beyond Figma (HN discussion, cited in research brief).
– Practitioner criticism of Figma Make's design-system fidelity ("ignore your components, skip your tokens," "every screen like a fresh snowflake") is drawn from real Hacker News discussion threads, not vendor marketing (HN discussion).
– On precise adoption numbers, market share, or a controlled benchmark comparing output quality across Figma Make, Claude Design, v0, and other named tools: evidence not sufficiently verified — the sources reviewed are practitioner commentary, a single blogger's direct comparison, and vendor-published comparison content, not an independent, controlled benchmark study.
– On the actual unit-economics claim (Figma paying for the same inference costs it competes against, versus smaller competitors' near-zero marginal cost): evidence not sufficiently verified as an audited financial comparison — this is analytical commentary from one source, not verified financial disclosure from either company.
Comparisons
Figma Make vs. Claude Design. Per the most detailed direct comparison found in this research, Figma Make is positioned as running on a comparatively weaker underlying model and lacking the same design-system import fidelity; Claude Design differentiates through onboarding-time design-system import (scanning codebase and design files for brand tokens, components, and voice) and a native handoff to Claude Code (Martin Alderson; Eigent.ai). 2026 comparison coverage frames the practical choice less as "which is better" and more as "which anchor fits your workflow" — Claude Design when your anchor is your design system/codebase, Figma Make when your anchor is an existing Figma workspace your team already lives in (Magic Patterns).
Figma Make vs. v0 by Vercel. Both are named among the competitive set explicitly discussed alongside Figma's AI stumbles; this research did not find a detailed feature-by-feature comparison between Figma Make and v0 specifically — treat any such comparison as evidence not sufficiently verified without checking current documentation from both directly.
Figma Make vs. Magic Patterns / Brilliant.design / Mowgli. Magic Patterns (YC W23) drew substantial Hacker News engagement (185 points, 114 comments) as a named alternative; Brilliant.design is described as working with vector primitives rather than code, positioning it as a distinct Figma alternative rather than an AI-layer-on-top-of-Figma approach; Mowgli is described as "Figma for the agent era" with Claude Code integration (HN discussion, cited in research brief; Brilliant.design; Mowgli) — each targets a meaningfully different niche rather than being interchangeable substitutes for each other.
Pricing: Claude Design vs. Figma Make. 2026 comparison coverage characterizes Claude Design's pricing as simpler and included within an existing Claude subscription, while Figma Make's usage-based pricing is described as more complex and capable of stacking up quickly under heavy use (Magic Patterns) — treat this as directional vendor-comparison commentary rather than a verified, itemized price comparison, since both companies' pricing structures can change.
Real-world use cases
– A team with an established design system evaluating AI tools for new feature mockups — the design-system-import capability, specifically credited to Claude Design, is the most directly relevant differentiator for this use case, since ignoring existing tokens creates rework rather than saving time.
– A designer already living inside Figma files who wants AI assistance without switching tools — Figma Make's point-and-edit refinement staying inside the existing Figma canvas is the specific advantage relevant here, despite the documented QA and fidelity criticisms.
– A non-designer founder needing professional-looking output fast — per the sourced comparison, this is specifically the scenario where a system-aware tool like Claude Design is credited with letting non-designers "create professional outputs matching corporate design standards."
– A product team handing off AI-generated designs directly to engineering — Claude Design's native handoff to Claude Code is the specific, sourced advantage for this workflow, versus staying purely inside a Figma-canvas-to-developer-handoff process.
Common mistakes
– Assuming any AI design tool automatically respects your existing component library and tokens. The core practitioner critique of Figma Make specifically warns against this assumption — verify against your own design system before relying on AI-generated output as production-ready.
– Treating the Apple Weather app incident as a one-off bug rather than a QA-process signal. Field's own public statement frames it as a QA-process failure, which is a more systemic concern than a single output glitch.
– Picking one AI design tool and assuming it covers the entire workflow. The emerging 2026 practitioner consensus is that most product teams end up using multiple tools for different stages (system-aware generation for one need, in-canvas point-and-edit refinement for another) rather than one tool for everything.
– Assuming AI-generated UI variety across different tools, when many default to the same shared component library (like shadcn). This can create a false sense of tool differentiation when the actual visual sameness comes from a shared underlying default, not from any one tool specifically.
– Treating vendor-published comparison content as neutral. Several of the more detailed comparisons available (e.g., from Magic Patterns, a competing product) have a commercial interest in how the comparison is framed — cross-reference against independent practitioner commentary (like the HN discussions cited here) rather than relying solely on vendor content.
Best practices
– Before adopting any AI design tool at a team with an existing design system, specifically test whether it respects your actual components and tokens, rather than assuming any tool does this by default.
– If your team already lives inside Figma files and values staying in that canvas for point-and-edit refinement, weigh that workflow continuity explicitly against the documented fidelity and QA criticisms of Figma Make.
– If your primary need is system-aware generation grounded in an existing codebase or design system, evaluate tools (like Claude Design) specifically on their design-system import mechanism, not just general output quality.
– Don't assume visually distinct output from two different tools reflects meaningfully different underlying capability — check whether both are simply defaulting to the same popular component library.
– Plan for using more than one tool across different stages of your design-to-development workflow, consistent with the 2026 practitioner pattern of pairing tools rather than picking a single winner.
– Weigh pricing structure carefully — a tool bundled into an existing subscription (as Claude Design is characterized) versus a tool with usage-based pricing that can scale unpredictably with heavy use (as Figma Make is characterized) has real budget-planning implications beyond the sticker price.
Key takeaways
– Figma's own "Make Designs" AI tool was disabled after a real, publicly acknowledged QA failure — it produced designs nearly identical to Apple's Weather app.
– The most consistent practitioner criticism of Figma Make is that it doesn't reliably respect existing design tokens and components, generating each screen as if it were a fresh, unrelated creation.
– Claude Design differentiates specifically through onboarding-time design-system import (scanning codebase and design files) and a native handoff to Claude Code — a materially different mechanism than Figma Make's context-frame approach.
– The economic argument that Figma pays for the same inference costs it competes against, while smaller rivals ship at near-zero marginal cost, is plausible analytical commentary, not verified financial data.
– The emerging 2026 practitioner pattern is pairing tools for different workflow stages (system-aware generation plus in-canvas refinement) rather than picking one AI design tool to do everything.
– Always verify an AI design tool's output against your actual design system and against well-known existing designs in your category before shipping, given the documented failure modes on both fronts.
Relevant tools.scult.in resources
If you're building or refining a brand's visual identity before feeding it into any AI design tool — as design-system context matters enormously for output quality per the criticism above — the Color Palette Generator and Favicon Generator are quick ways to lock down foundational brand assets first. For prompt patterns specifically for working with Claude or ChatGPT on design tasks, the Claude and ChatGPT prompt libraries have relevant starting points.
If your team needs a coherent, properly documented design system built or cleaned up before any AI design tool can meaningfully respect it — since, as this article covers, garbage-in-garbage-out applies directly to design-system fidelity — that foundational design work is exactly where SCULT's branding agency team can help, so whichever AI tool you adopt afterward has something real to import.
Frequently asked questions
Why did Figma disable its AI design tool?
It produced app designs nearly identical to Apple's Weather app, and CEO Dylan Field publicly attributed this to an insufficient QA process (404 Media).
What is Figma Make?
Figma's AI feature for prompt-to-prototype generation that stays editable directly on the Figma canvas, and can use existing Figma frames or "Make kits" as design-system context (Figma).
What is Claude Design?
Anthropic's AI design tool, differentiated by importing an existing design system (codebase and design files) during onboarding and applying it to future generations.
Is Figma Make good enough to replace a dedicated AI design tool?
Practitioner feedback is largely skeptical — one commenter called it "incredibly underwhelming," and criticism centers on it ignoring existing components and design tokens (HN discussion).
Does Figma Make respect my design tokens?
Practitioner reports suggest not reliably — criticism specifically describes it treating "every screen like a fresh snowflake" rather than reusing existing tokens and components.
What happened to Figma's AI design tool?
It was disabled after producing output nearly identical to a well-known existing app (Apple's Weather app), pending further QA work (404 Media).
Is Claude Design better than Figma Make?
Per the most detailed direct comparison found in this research, Claude Design is argued to run on a stronger underlying model and produce more on-brand results via design-system import — though the two are increasingly framed as fitting different workflow anchors rather than one being categorically "better" (Martin Alderson; Magic Patterns).
What other AI design tools compete with Figma's built-in AI?
v0 by Vercel, Magic Patterns, Brilliant.design, and Mowgli are all named alternatives discussed alongside Figma's AI stumbles.
Can non-designers get good results from an AI design tool?
Yes, per at least one sourced account specifically about Claude Design, which is credited with letting non-designers produce output matching corporate design standards via one-click design-system import (Martin Alderson).
Do I have to choose just one AI design tool?
No — the emerging 2026 practitioner pattern is using multiple tools for different stages of the same workflow rather than picking a single tool for everything (Magic Patterns).
What specifically went wrong with Figma's "Make Designs" tool?
It generated app designs nearly identical to Apple's Weather app; Field attributed the failure to insufficient QA process rather than a one-off fluke (404 Media).
Why do practitioners say Figma Make ignores components?
Reported firsthand practitioner criticism describes it generating pages that skip existing tokens and components, treating each screen as a fresh generation rather than reusing the team's actual design system (HN discussion).
How does Claude Design's design-system import actually work?
During onboarding, it scans a team's codebase and design files to learn brand tokens, components, and voice, then applies that learned system to future generations (Eigent.ai).
Why does design-token fidelity matter so much for evaluating these tools?
Because ignoring existing tokens/components doesn't just produce lower-quality output — it creates rework, since someone has to manually reconcile AI-generated screens with the team's real standards afterward.
What does "every screen like a fresh snowflake" mean as a criticism?
That the tool isn't reusing or building on an established design system — each generated screen is treated as an independent creation rather than a consistent extension of prior work.
Does Figma Make use its own proprietary model?
One practitioner account describes it as "basically just Gemini 3 Pro with a design prompt," suggesting the underlying model may not be a Figma-specific, purpose-built model — though this is a practitioner characterization, not confirmed technical detail from Figma itself.
What does Claude Design's native handoff to Claude Code actually enable?
It lets an engineering team pick up and continue generated design work directly through Claude Code, tying design generation more closely to implementation than a Figma-canvas-only workflow (Eigent.ai).
Why do many AI-generated UIs look similar regardless of the tool used?
One commenter attributed this to many teams and tools defaulting to the same popular off-the-shelf component library (shadcn specifically named) as a shared starting point.
Did Figma's AI stumble create an opening for competitors?
Commentary frames it as a structural disadvantage for Figma specifically — Figma pays for the same AI inference it's competing against, while smaller dedicated competitors can ship at near-zero marginal cost (Martin Alderson) — though this is analytical commentary, not audited financial data.
Is the criticism of Figma Make specific to one bad release, or an ongoing pattern?
Both — the Apple Weather app incident is one specific, dated event, but the design-system-fidelity criticism from practitioner discussion describes an ongoing pattern in the tool's general behavior, not just that one incident.
How do I use Figma Make with my existing design system?
Provide existing Figma frames or "Make kits" as context during generation, per Figma's own documented approach — though verify the output actually respects your tokens and components given the sourced criticism above.
How do I prompt Figma AI to match my components?
Reference specific component names and provide existing frames as direct context; practitioner reports suggest this doesn't reliably guarantee fidelity, so manual review against your system remains necessary.
How do I evaluate whether an AI design tool respects my design system before committing to it?
Run a test generation using your actual design system as input and check whether the output reuses your specific tokens and components, rather than trusting a vendor's general capability claims.
How do I get Claude Design to learn my existing design system?
Through its onboarding process, which scans your codebase and design files to learn brand tokens, components, and voice automatically (Eigent.ai).
How do I hand off AI-generated designs to engineering efficiently?
If using Claude Design, its native handoff to Claude Code is a documented path; if using Figma Make, the handoff stays within Figma's existing design-to-development tooling.
How do I avoid generic, sameness-prone AI-generated UI output?
Be aware that shared default component libraries (like shadcn) can make output from different tools look similar regardless of the tool used — explicitly customizing beyond defaults is necessary either way.
How do I compare Figma Make and Claude Design for my specific team?
Weigh whether your primary anchor is an existing Figma workspace (favoring Figma Make's in-canvas workflow) or an existing design system/codebase (favoring Claude Design's import mechanism) (Magic Patterns).
How do I decide whether to trust an AI design tool's output for production use?
Cross-check against your actual component library and, given the documented Apple Weather app incident, specifically check for unintended similarity to existing well-known designs.
How do I combine multiple AI design tools in one workflow?
Follow the emerging 2026 practitioner pattern: use a system-aware tool for on-brand generation and a canvas-based tool for point-and-edit refinement, rather than expecting one tool to do both well.
How do I evaluate newer, smaller AI design tools like Magic Patterns, Brilliant.design, or Mowgli?
Check what specific niche each targets — Magic Patterns and Brilliant.design position differently around code-vs-vector-primitive generation, and Mowgli specifically integrates with Claude Code — rather than assuming they're interchangeable general-purpose competitors to Figma Make.
Does Figma Make's underlying model differ from Claude Design's?
Per one practitioner characterization, Figma Make may run on Gemini 3 Pro with a design-specific prompt layer, while Claude Design is built on Anthropic's own models — this is a practitioner claim, not independently verified technical confirmation from either company.
Is there a technical reason design-system import (Claude Design's approach) produces more reliable brand fidelity than context-frame input (Figma Make's approach)?
The sourced comparison argues yes, attributing it to a more thorough onboarding-time system-learning process versus per-generation context frames, but this wasn't independently benchmarked with a controlled technical evaluation in this research.
How does Brilliant.design's "vector primitives, not code" approach differ mechanically from code-generating tools like v0 or Claude Design?
Working with vector primitives directly, per its own positioning, keeps output in a design-native format rather than generating code as the primary artifact — a meaningfully different technical approach, though this research did not verify deeper implementation details beyond that positioning.
What does "Figma for the agent era" mean as Mowgli's positioning?
Per its own description, a canvas-style design tool built specifically for AI-agent-driven workflows with Claude Code integration, distinguishing it from both Figma's own AI features and from code-first tools like v0.
Is the Figma-pays-for-inference-it-competes-against economic argument verifiable?
Not with audited financial data in the sources reviewed — evidence not sufficiently verified; treat it as a plausible structural explanation from one analytical source, not confirmed unit economics.
Figma Make vs. Claude Design — which should I use?
Depends on your anchor: Claude Design if your priority is design-system fidelity and codebase-grounded generation; Figma Make if your priority is staying inside an existing Figma workspace with point-and-edit refinement (Magic Patterns).
Figma Make vs. v0 by Vercel — how do they compare?
Both are named in the same competitive conversation, but this research did not find a detailed direct feature comparison between them specifically — evidence not sufficiently verified beyond their shared framing as AI design/prototyping alternatives.
Claude Design vs. Magic Patterns — what's the difference?
Not directly compared in the sources reviewed — evidence not sufficiently verified; Magic Patterns is documented mainly as a named, well-discussed alternative (185 points/114 comments on HN) rather than benchmarked directly against Claude Design.
Figma's own AI vs. a Figma alternative like Brilliant.design — which is the bigger shift?
Brilliant.design represents a more fundamental alternative (a different tool entirely, working with vector primitives rather than code) than Figma Make, which is an AI feature layered onto the existing Figma product.
Vendor-published comparisons (e.g., from Magic Patterns) vs. independent practitioner commentary (HN) — which should I trust more?
Independent practitioner commentary generally carries less commercial bias, though it's also less systematic; the most reliable approach is cross-referencing both rather than relying on either alone.
My AI-generated design looks generic and doesn't match our brand — what's likely wrong?
Likely a design-system-fidelity issue — check whether the tool actually ingested and applied your existing tokens/components, or defaulted to a generic template/shared component library.
I'm worried an AI design tool might produce output too similar to an existing app — how do I check?
Given the documented Apple Weather app incident, explicitly review generated output against well-known existing designs in your category before shipping, rather than assuming originality by default.
Figma Make ignored the components I referenced — why?
Consistent with the widely reported practitioner criticism that it doesn't reliably respect existing components and tokens even when provided as context — manual reconciliation may currently be necessary.
We adopted an AI design tool and now everything looks like every other AI-generated site — what happened?
Likely a shared-defaults issue (many tools lean on the same popular component library, like shadcn) rather than a problem specific to your chosen tool.
Our design and engineering teams are struggling with AI-generated design handoff — what should we check?
Whether your chosen tool has a native handoff path (like Claude Design's to Claude Code) or whether you're relying on a manual export/translation step that's introducing friction.
Should we pay for Claude Design, Figma Make, or both?
The 2026 practitioner pattern suggests most teams benefit from using both for different stages — evaluate based on your specific anchor (design-system fidelity vs. in-canvas Figma workflow) rather than assuming an either/or choice.
Is Figma Make worth using despite the documented QA incident and fidelity criticism?
Worth evaluating for teams that specifically value staying inside an existing Figma workflow for point-and-edit refinement — but verify design-system fidelity yourself given the sourced criticism, rather than assuming it's been fully resolved.
Is Claude Design worth adopting for a team without any existing design system?
The sourced account suggests yes for non-designers specifically, since it's credited with helping produce professional, corporate-standard output even without an extensive existing system to import from.
How should pricing factor into choosing between these tools?
Claude Design's pricing is characterized as simpler and bundled into an existing subscription; Figma Make's is characterized as usage-based and able to scale unpredictably under heavy use — weigh this against your expected generation volume, though verify current pricing directly with each vendor.
Should a startup without a dedicated designer invest in one of these tools now, or wait for the category to mature further?
Given both the real capability gains documented (professional output for non-designers) and the real documented risks (the Apple Weather app incident, fidelity criticism), a cautious pilot with explicit human review of output — rather than either full adoption or complete avoidance — is the best-supported approach from the evidence reviewed.
Sources
- https://www.404media.co/figma-disables-ai-app-design-tool-after-it-copied-apples-weather-app/
- https://www.figma.com/make/
- https://martinalderson.com/posts/figmas-woes-compound-with-claude-design/
- https://hn.algolia.com/api/v1/search?query=Claude%20Design%20Figma%20Make
- https://www.eigent.ai/blog/claude-design-vs-figma-make
- https://www.magicpatterns.com/blog/claude-design-vs-figma-make
- https://www.lennysnewsletter.com/p/what-claude-design-is-actually-good
Need this built into your business?
The free tools and prompts on this site handle the small, solved problems. If what you need is bigger — branding & design, built and maintained for you — that's Scult's day job.
