Can abstract gradients build trust in AI systems better than icons?

Google Design's account of building Gemini's illustration system frames the central design problem as trust: "How do you build trust in a tool that's always evolving?" The designers explicitly reject two prior answers to that problem. One is Susan Kare's object-based Macintosh icons — "a trash can, a paintbrush, a smiling computer face" — which worked as "bridges between human understanding and machine logic" precisely because they depicted things. The other is a literal assistant character, "unlike the static nature of past digital assistants (many of us remember Microsoft's Clippy)." Gemini's team instead "landed on gradients," which are "much more about energy than 'objectness,' like Kare's illustrations (a trash can is a thing, a gradient is a vibe)." Design lead Anna Sera Garcia describes the resulting look as having a "warm, spatial, rounded quality," with an "ethereal" and "fuzzy space that reflects our nonlinear process for ideation."

The mechanism is built from two elements. Circles ground the system in Google's existing brand — Gemini's own logo is "constructed from the negative space of four adjoining circles," and buttons and containers echo the shape through rounded corners, while the historic four-color Google dots and softened, blurred Material shapes supply "a familiar history" that the team evolves into something more fluid. Gradients then do the expressive work: they act as "context builders" with "sharp, almost opaque leading edges that diffuse at the tail," directing attention and conveying "a transfer of energy and directional momentum." Motion completes the system — "each animation has a defined start and end point," mirroring user actions so that "inner activity within the motion conveys thinking, analysis, and intelligence," making Gemini's processing "feel more transparent" without requiring users to see any literal mechanism.

This sits beside two other accounts of how AI systems earn trust through what users can see. Does Gemini agent truly maintain continuous memory across devices? describes the same product building trust through persistent memory continuity — users don't need to re-brief the agent — while this source builds trust through moment-to-moment visual signaling of the agent's internal state. It also contrasts with How should AI interfaces handle the shift from doing to supervising?, which locates AI trust-building in explicit surfaces where users state intent or negotiate plans; Gemini's illustration team instead works in an ambient, affective register beneath those surfaces, signaling "thinking" without exposing anything to negotiate. It also differs sharply from Can a shared canvas serve both human and agent memory?, where agent state is a structured object users can inspect and manipulate directly; Gemini's gradients gesture at process atmospherically but give the user no lever on what is actually happening.

The excerpt is a design team's own retrospective and promotional account, not a study: it offers no user testing, no comparison of how gradients perform against Kare-style icons or a mascot on actual trust or comprehension measures, and no evidence that feeling more transparent corresponds to users understanding or controlling anything about Gemini's behavior. The claim worth taking away is narrower than it sounds — that one major AI product deliberately chose abstraction over personification or literal iconography as its design bet for conveying "thinking" — not that this bet has been shown to work.