Figma pulls its AI tool after copying Apple’s Weather app

Ask an AI to draft a fresh weather app, and it hands you an exact replica of Apple's iOS interface down to the card layout.

That is the exact mess Figma found itself in this week. On Tuesday, Figma CEO Dylan Field announced that the company is temporarily pulling the plug on its brand-new "Make Design" AI feature after users caught it repeatedly cloning Apple’s Weather app.

A deadline rush ends in a clone

The issue was flagged by Andy Allen, founder of Not Boring Software. While testing Figma’s new generative feature, Allen noticed that prompts asking for a weather app consistently spat out designs virtually identical to the native Apple Weather app.

Allen took to X to share screenshots and warned fellow designers to tread carefully. He openly accused Figma of heavily training its AI on existing app designs, pointing out that using these raw outputs in production could easily land designers in serious legal trouble with copyright holders.

Make Design was the marquee feature unveiled at Figma’s Config conference just last week. The premise was simple: describe what you want in plain text, and the AI drafts a clean set of UI layouts and components to kickstart your project.

Instead, it took less than a week for the flagship tool to be turned off.

Under the hood of the blunder

Field was quick to respond on X, pushing back hard on the data theft accusations. He stated that Figma did not train the models on user content, community files, or existing app designs, calling those claims flat-out false.

According to Field, Make Design relies on off-the-shelf third-party large language models paired with custom design systems commissioned by Figma. The root cause of the uncanny replication was not model memorization, but rather that the commissioned design systems had extremely low variability.

In plain English: the guardrails and component recipes fed to the LLM were so rigid that the AI predictably assembled the most generic, universally known weather layout imaginable.

Field took personal responsibility for the rollout, admitting he pushed the engineering team too hard to meet the Config deadline and compromised on standard QA processes. The feature is disabled until Figma finishes a complete overhaul and testing pass on the underlying system.

Why it matters

This incident exposes the messy reality behind text-to-UI generative tools.

For designers, this is an immediate reality check. The promise of skipping wireframing with a single text prompt sounds incredible, but if the output is an inadvertent replica of proprietary software, it is an absolute nonstarter for serious commercial work. Nobody wants to explain to their legal team why their new startup looks like an Apple asset pack.

We think Figma did the only sensible thing by benching the feature immediately, but rushing it out for a conference keynote was a clumsy mistake. It reveals how software companies are cutting corners on basic quality control just to boast an AI story to investors and users.

Looking at the bigger picture, using off-the-shelf LLMs to arrange pre-baked UI components is inherently fragile. Without vast, carefully tuned design systems with high variance, these models will inevitably converge on the most prominent design patterns in tech.

Until tooling platforms can guarantee genuine output diversity and zero infringement liability, AI-generated layouts are going to remain a risky novelty rather than a reliable production asset.