Figma Agent accepts prompts and generates content directly on the canvas, with access via a sidebar and a star icon on selected elements. Early tests focused on core design tasks. When asked to generate a missing pawn in a minimal chess set based on the existing style, the output was a buggy drawing that failed to match the original visual qualities. After adding a homemade pawn, a request to arrange pieces into a chess board also failed, producing an incorrect 3x3 board full of bugs. The results suggest generative AI struggles with meaningful icon-like work, though vector outputs could still enable human refinement.
"Figma will now take your prompts and generate directly on the canvas. Is this the tipping point for AI in design? The basics If you have beta access, there are two entry points for letting Figma design for you: the first one is through a sidebar, much like other vibe coding tools out there. In addition to this, there’s a star icon whenever you select something, letting you fire off prompts directly from there."
"I was working on some sketches of a minimal chess piece set, and deliberately left out the pawn for the agent to generate. I selected the five remaining pieces and asked the agent to generate a pawn based on the style. Unfortunately, the result was nothing more than a waste of power, credits and patience on my end: A buggy drawing that completely missed the visual qualities of the original set."
"After dismissing this and adding my homemade pawn, I asked the agent to order them nicely into a chess board. A chore that would actually be nice to offload, but this too failed badly, returning a 3x3 board full of bugs. I'm not overly surprised: generative AI seem to struggle to do meaningful work in the realm of icons."
"There is however some promise in getting vector designs back, opening the door for tweaks and further refinements by a human. But all in all, this first attempt was nowhere near helpful."
Read at Shapes by Gjermund Gustavsen
Unable to calculate read time
Collection
[
|
...
]