Maximillian Piras

Maximillian Piras is the founding designer at Yutori, where he shapes interfaces for software agents that navigate websites and complete tasks. His work addresses a central problem of AI product design: how to make increasingly capable systems understandable without locking them into interaction patterns that quickly become obsolete.

Piras’s portfolio credits design projects for the music platform 8tracks, and his writing describes onboarding work at Headliner. His creative portfolio spans product interfaces, animation, illustration, commissions for Giphy, and a music-video project involving Ryuichi Sakamoto. His Headliner onboarding work captured individual users’ preferences; he reports no meaningful activation dropoff as onboarding steps were added, alongside improved retention from personalization.

In a 2023 essay on machine-learning interfaces, Piras argued that strategic interface friction can improve personalization, reduce mistakes, and generate clearer behavioral signals. Extra onboarding questions proved valuable when they produced more useful recommendations without the abandonment his team had anticipated. His 2024 writing on alternatives to conversational AI extended that thinking to language models: visual manipulation, navigation, and comparison can outperform chat when matched to the task.

• The Bitter Layout: Piras’s analysis of convergent AI interfaces explains why text input, sequential conversation, and model selectors keep reappearing. While model capabilities remain a competitive differentiator, a generic interface can absorb improvements faster than specialized workflows built around yesterday’s limitations.

• The flexibility-usability tradeoff: Interfaces optimized for well-understood tasks can be clearer and more efficient, while general-purpose systems accommodate unexpected tasks and rapidly changing models. For Piras, the right balance depends on a product’s users, technical constraints, and stage of development.

• Model pickers as interface modes: Switching models can silently change available features, tools, and output quality. When users must match a model to a compatible feature, they inherit the product’s internal complexity instead of concentrating on their goal.

• Designing with goals and constraints: As AI systems become less predictable, Piras sees designers specifying boundaries and desired outcomes instead of scripting every possible interaction. Potential approaches include using design systems to constrain generated interfaces and translating user stories into higher-level instructions.

Piras tests AI interfaces beyond chat through CC-1, an LLM-powered word-arithmetic calculator.