Skip to content
Article Figma Blog Sep 2026

Figma Blog: What it takes to build great products now

Paige Costello, VP of Product at Figma, published a short essay in September 2026 asking what has changed—and what has not—for product designers now that AI can generate a working prototype in minutes. The piece is grounded in her observations of teams actually shipping with these tools, not in predictions.

The central tension Costello names is this: speed to first prototype has never been higher, but the gap between a first prototype and a great product has not gotten smaller. AI tools are exceptional at getting you started. They are far less useful at knowing when to stop, when to explore further, or when to make a deliberate choice that gives a product a distinct character. Those are the skills she argues designers need to develop now, not because AI is weak in general, but because this specific gap is where competitive differentiation is happening.

Costello describes three areas where human design skill still matters most.

The first is sustaining momentum through the final stages of a project. Early phases are well served by generative tools—you get options quickly, the surface area of what’s possible expands fast. But late-stage design work, where you are resolving edge cases, making the product feel finished, and verifying that the thing actually works for real users in real contexts, still requires sustained human attention. The tendency to call something done when the AI-generated version looks good enough is a pattern she flags explicitly.

The second is exploring multiple directions before committing. Costello argues that AI makes it tempting to converge quickly on the first plausible solution, because the first plausible solution is now extremely quick to produce. Designers who deliberately generate and evaluate three or four meaningfully different directions before committing will outperform those who optimize the first result they see. The skill is not ideation—it’s knowing how to stay in exploration long enough to find something better.

The third is building with distinctiveness deliberately. Products that feel interchangeable tend to be built by teams that accepted AI defaults without interrogating them. Visual language, interaction style, and the way a product speaks to its users require intentional choices that go against the median output of a generation model. Costello frames this as a craft question: you are responsible for what makes your product recognizably itself.

The essay also previews several Figma features—Make, Motion, Draw, and Weave—as tools designed to support these later-stage and more intentional design activities. The framing is useful context rather than a product announcement, though readers should note the authorial perspective.

This piece is worth reading for designers who are already using AI generation tools and are starting to feel the gap Costello is describing—the point where the prototype looks fine but the product is not quite there. It is also useful for design leads thinking about how to frame skill development for their teams. The argument is short enough to circulate and concrete enough to act on.