How BUCK gave Notion’s AI assistant a hand-drawn personality

AI assistants have developed a visual language. Sparkles suggest magic. Asterisks hint at computational intelligence. Gradients imply futuristic capability. Notion looked at this crowded landscape and chose a different path entirely: a simple illustrated face with eyes, eyebrows and a nose that changes expression based on what it’s doing. The character emerged from Art Director Roman Muradov’s signature illustration style. Fast Company’s Mark Wilson captured its personality precisely: “It looks like Microsoft’s Clippy grew up drinking dirty martinis and reading New Yorker cartoons.” That comparison to Microsoft’s infamous assistant isn’t accidental – both attempt to humanise software through character, but where Clippy became intrusive, Notion AI aims for something more subdued. BUCK received the challenge of animating Muradov’s static drawings into a dynamic character that would respond to user actions across Notion’s millions of users. The brief carried a specific constraint: Notion AI needed to feel delightful – alert and engaged, but not distracting. This tension between personality and restraint shaped every animation decision that followed. The production began with cel animation, a deliberate choice to preserve the handmade quality of Muradov’s original drawings. With just three facial elements to work with, BUCK’s animation team developed distinct behaviours for each interface state. Eyebrows wave whilst thinking. The face momentarily falls apart during errors. Each expression needed to feel recognisably human whilst remaining within the minimal design language Notion had established. The technical challenge went beyond individual animations. The product demanded complete fluidity – the face needed to transition seamlessly between any state, adapting to user interactions in real time. A character might shift from listening to thinking to writing within seconds, and each transition needed to feel organic rather than jarring. Traditional animation pipelines couldn’t accommodate this level of responsiveness without file sizes that would impact performance.