This week, we’re detailing Figma’s eval process for its new AI product, Figma Make — one that centered around scaling human judgement.
The Art of Evals: How Figma Put People at the Center of Its AI Product

Building an AI product requires a fundamentally different process. Unlike traditional software where there’s a clear path to see what’s possible, the capabilities of an AI product exist in a foggy middle ground that’s only validated through actual user testing.
“There’s a different playbook being developed inside every product team right now about how to scale human taste,” says David Kossnick, Figma’s Head of Product, AI. And to create its new prompt-to-functional-app experience, Figma put human craft and creativity at the center — from defining success metrics, to the process for gathering qualitative feedback, to exploring how to assess and structure this data.
In this essay, you can expect to learn more about:
- Figma’s decision tree for assessing AI product viability — the four paths Kossnick considers before investing further into any AI product, depending on current technology and if you can spend time making adjustments to your product scope or features.
- Tips for constructing an AI team — AI tools can make product teams smaller and more nimble, which is extremely valuable as prototypes enable faster iteration (just don’t forget to include the target persona on the team).
- The importance of prototyping — which Kossnick says is the new gold standard as a validation mechanism. He and the team got an “unoptimized prototype” to users as fast as possible to see what they did with the product and if it met their expectations, which shaped its development.
- How to think about measuring success — especially when there are multiple facets of success and you need to determine which is the right one for your product.
- The concentric circles of feedback Figma used for evals — starting small and getting broader, but always making sure the product was measured against the expected quality of its outputs.
The ground is constantly shifting under those who are building, testing and validating AI products. As new technologies and standards emerge, thinking at a higher level about how to approach these challenges can give you a way to navigate that ever-present movement.
Thanks, as always, for reading and sharing.
-The Review Editors