Figma has shipped a substantial update that brings new tools for designers and developers to its collaborative platform. Among the highlights are a dedicated code layer, native support for animations and shaders, and the ability to create custom plugins that leverage AI models for specific tasks. This marks another step in weaving AI into UI/UX workflows, but it also reignites the conversation about where computation and data actually reside.

Code and motion become first-class layers

The “code layers” feature lets teams attach code snippets directly to canvas elements, bridging the gap between prototype and implementation. It goes beyond static annotations: developers can see in real time how code impacts the UI, reducing cross-team friction. At the same time, motion and shader primitives enable complex transitions and visual effects without leaving Figma, speeding up iteration on dynamic experiences.

AI arrives in plugins, but the boundary is the cloud

The third pillar of the announcement is the densest: the platform now supports plugins that integrate AI models, allowing advanced users to automate tasks like asset generation, text translation, or layout analysis. This is where the update grazes the sensibilities of those who follow the evolution of on-premise models. Figma is a cloud-native service, and all its AI capabilities—including those accessible through plugins—run on vendor-managed infrastructure. Companies that, due to compliance constraints (GDPR, industry regulations, security audits), must keep design projects inside their own data centers are effectively cut off from directly using these features.

Digital sovereignty and trade-offs

From an AI-RADAR perspective, which examines self-hosted LLMs and local stacks, the Figma news serves as a relevant case study: on one hand, cloud-first tools like Figma accelerate AI adoption in design by democratizing advanced functionality; on the other, anyone pursuing data sovereignty faces the choice of either sticking to external platforms or building internal pipelines based on self-hosted frameworks, at higher development and operational costs. There’s no one-to-one mapping—Figma is not an LLM—but the dynamics mirror those encountered when weighing a shift from cloud services to local inference: the loss of certain integrated features in exchange for full control over data.

A signal for the industry

The update confirms the trajectory of professional creative tools: AI is becoming an infrastructural component, not an accessory. While vendors like Figma push the frontier toward automation and assisted generation, those working on on-premise deployments know that the real value lies in orchestrating these capabilities without surrendering data residency. The fledgling AI plugin ecosystem might one day enable hybrid or client-side execution, but for now the message is clear: rapid innovation in the cloud, caution on-premise. For teams exploring alternative paths, AI-RADAR offers analytical frameworks to measure the trade-off between TCO, performance, and control.