With new AI features landing in Figma seemingly every quarter, many teams are struggling to tell which ones are genuinely ready for production work and which are still closer to demo material. For any team considering Figma AI, the first order of business — before a single license fee is paid — is getting a clear picture of what actually works and what doesn’t. Decide based on a polished demo reel alone, and you are likely to be disappointed by the gap between expectation and reality once the tools meet real projects.
1. What Works Today, and What Doesn’t Yet
What Figma AI can already do for you
Several features are already saving real time in day-to-day work. Searching for layers and components with plain-language text queries, automatically removing backgrounds from images, and filling designs with placeholder text or mock content all deliver on their promise. The feature that infers connections between screens and wires up a prototype automatically is genuinely useful in early wireframing, when speed matters more than precision. And the text summarizing and rewriting tools are perfectly serviceable for roughing out a first pass at copy.
Where the limits are still obvious
What Figma AI cannot yet do is produce a polished, finished UI from a single prompt. Generated output frequently ignores the rules of your team’s design system — component naming conventions, auto layout structure, color tokens — which means a designer ends up manually reorganizing what the AI produced. Complex interactions and unusual responsive rules are another weak spot: the AI often fails to grasp them and improvises instead. The underlying reason for these limits is structural. AI is good at recognizing the visual patterns that appear over and over on screens, but it does not understand the implicit judgment behind those screens — brand context, business constraints, accessibility requirements. Imitating a pattern and understanding an intent are problems of a different order.
Working today
Layer and component search, automatic background removal, mock content fills, inferred prototype connections, text summarizing and rewriting.
Still clearly limited
Polished UI from a one-line prompt, automatic compliance with design system rules, complex interactions and responsive behavior.
2. What to Sort Out Before Adopting AI — the Design System Is the Foundation
A common misconception in practice is that once AI arrives, the design system cleanup a team has been postponing can finally be skipped. The truth is closer to the opposite. For text-based layer and component search to work properly, naming conventions must be consistent in the first place. In a file where component names are all over the place and auto layout structures are a tangle, the AI will simply return an equally tangled result. Put plainly: AI is a tool that lets you exploit an existing structure faster — it is not a tool that builds the structure for you.
Confirm that component and layer naming conventions are applied consistently across the whole team
Check that your design token system — color, typography, and the rest — is in order
Review whether your auto layout structures are standardized
If things are messy, do this foundation work first, before adopting AI
3. Experimental Territory: Features Like AI Shaders
The traditional workflow was for a designer to hunt down references by hand and build mockups manually to compare directions. It was slow, but it had one great virtue: the designer’s intent was carried through to the smallest detail. The AI-assisted workflow dramatically shortens the exploratory phase at the front of that process — with the standing caveat that the detailed judgment calls still need a human review afterward. More recently, a workflow has emerged around using Figma’s AI shader features to produce CMF concepts — color, material, and finish. Being able to adjust texture and surface quality through prompts and compare multiple directions quickly makes this especially useful in industrial design; we covered the workflow in detail in our guide to building CMF concepts with Figma AI shaders. That said, features like these are still experimental, and the safe posture is to treat them as tools for exploring direction rather than for producing final deliverables. The more experimental a feature is, the more its behavior can change from one version to the next — so rather than trusting a workflow you learned once, make a habit of revisiting it periodically.
💡 Pro tip — Everything the AI generates must be re-reviewed against your component rules and design tokens. Having the token system sorted out in advance makes that review dramatically easier. If you don’t have a token structure yet, start with our primer on what design tokens are.
4. What to Check Before Putting It Into Production
Have you confirmed which AI features your current plan actually includes?
Is there a review step that checks AI-generated output against your component rules?
Are experimental features like AI shaders being used for exploration, not final deliverables?
Are naming conventions and the token system in order before AI is layered on top?
Closing thoughts
Figma’s AI lineup keeps evolving — the official Figma AI page is updated continuously, and it is worth re-running this works-versus-doesn’t assessment each time something new ships. The conclusion, for now, is simple: at this stage AI is an assistant that trims away repetitive work. It is not a substitute for a designer’s judgment.
Design Daily Life · Notes on design, daily