I’ve been watching the UI design tools space like a hawk for the past three years, and I’ve seen a clear pattern. Phase one: you throw a prompt into Midjourney or DALL·E, wait a minute, get four random images, pick the least ugly one, and then spend another ten minutes trying to regenerate a detail without breaking everything else. Phase two: someone built an Agent wrapper on top of those models—a chat interface that “understands” your intent, but the agent is still a plugin floating above the canvas. You talk to it, it generates something, but you’re still in two separate worlds: the chat log and the actual design file.
Then I tried out Claude’s new “Design” feature a few days ago. Opened it up, and the first thing that hit me was: this is not a plugin. This is the canvas.
Let me unpack that for a second.
Figma is a brilliant piece of software. Seriously, no sarcasm. It killed Sketch and XD by being web-native and multi-player. But its core assumption is that you are the designer. You move pixels, align layers, adjust spacing. The tool is a perfect mirror of your hands. Fast forward to 2025, and that assumption is suddenly the weakest link in the chain.
Claude Design works differently. You describe what you want—something like “a mobile checkout screen with Apple Pay, a promo code field, and clear error states”—and Claude draws the fucking thing. Right there, inside a canvas that feels alive. It’s not a “generate and paste” flow. It’s a conversation where the canvas updates in real time as you talk. You can click on any element, ask Claude to tweak it, and the canvas morphs. No more “export to Figma, then adjust”. No more “take a screenshot, paste into chat, wait for feedback”.
This changes everything about how design gets done.
Now, a lot of people will say: “Figma will just add AI features, duh.” And sure, Figma has their AI stuff. But the thing is—and this is the key insight that gets missed—Figma’s DNA is built around a manual tool paradigm. Any AI feature they add is going to be a band-aid on top of a tool designed for manual pixel-pushing. It’s like attaching a rocket engine to a horse carriage. You can do it, but the fundamental architecture fights you every step of the way.
Claude Design isn’t just “Figma with AI”. It’s a new genre of tool. The canvas is not a static representation of your work; it’s a live state machine driven by language. The designer’s job shifts from operating the tool to specifying the intent. And that shift is fucking huge.
Look at the history of creative tools. Photoshop killed the darkroom. Figma killed Photoshop (for UI, at least). Each new tool didn’t just do the old job faster; it enabled a different way of thinking. Photoshop let you undo and layer. Figma let you collaborate in real time. Claude Design lets you talk to the design.
And here’s where it gets uncomfortable for the design world: when the tool becomes that fluent, a lot of what we call “design work”—the pixel-pushing, the alignments, the spacing—becomes meaningless busywork. The real value moves upstream: understanding the user problem, deciding the information architecture, crafting the copy that guides behavior.
Figma never competed on those things. It competed on being the best pixel canvas. But the whole concept of a “pixel canvas” is being dissolved by LLMs that can generate pixel-perfect UI from text faster than any human can drag a mouse.
So, yes, Claude Design is killing Figma. But it’s not because Figma is a bad product. It’s because the whole category of “design tool” is being redefined. Figma was the king of the file-based canvas era. Claude is the harbinger of the intent-based design era. You can’t bolt an LLM onto Figma and call it a day. You need to rebuild from the ground up with language as the core primitive.
And that’s why this whole debate is so frustrating to watch. People argue about features, about pricing, about Figma’s AI roadmap. But the real story is a paradigm shift that makes all those arguments irrelevant. It’s like arguing about whether the Model T has a good horse-comparison chart. You’re missing the point.
The next time you’re in a design tool, try this: ask yourself, “Am I thinking about the design, or am I thinking about how to use the tool to express the design?” If you find yourself doing the latter, you’re on borrowed time. The tools that let you think about the design will win. And Claude Design, right now, is the first tool I’ve seen that actually delivers on that promise.
Not because it’s perfect—it has rough edges, it hallucinates, it can’t handle insane pixel precision (yet). But because its fundamental architecture is aligned with where we’re heading.
Figma’s problem isn’t that it’s bad. It’s that the rules of the game have changed. And nobody told the referee.