Unlike traditional deterministic products, AI models are probabilistic; the same query can yield different results. This uncertainty requires designers, PMs, and engineers to align on flexible expectations rather than fixed workflows, fundamentally changing the nature of collaboration.

In an AI-driven environment, interfaces are often conversational and nondeterministic rather than static step-by-step UI flows. Francois Lopitaux argues that evaluation systems (evals) become the new PRD. Because product managers cannot dictate the exact UI paths users take, evals are essential to define and validate expected behavior, score accuracy and consistency, and establish clear guardrails for what outputs are acceptable versus unacceptable.

At OpenAI, the first question is "Can we solve this with the model (tokens) instead of pixels?" This treats the AI as the primary design material, pushing designers to think about interaction and behavior before creating bespoke user interfaces.

For non-deterministic AI systems where every output can't be controlled, designers' roles must evolve to creating "evals"—systems that define and test for quality. Figma's CDO argues this is crucial for ensuring experiences generated by algorithms or LLMs consistently meet the desired standard of "good."

Giving AI freedom to generate custom layouts dynamically risks disorienting users if interface patterns shift with every response. Rhiannon Bell explains that while AI can assemble components from design systems, designers must enforce human-defined rules. Predictability in where answers, interactive widgets, and links appear is critical so users do not have to learn a brand-new interface on each query.

As AI models become proficient at generating high-quality UI from prompts, the value of manual design execution will diminish. A professional designer's key differentiator will become their ability to build the underlying, unique component libraries and design systems that AI will use to create those UIs.

OpenAI is developing a "dynamic user interface library" designed so the AI model can interpret and compose UI elements itself. This forward-thinking approach anticipates a future where the model assembles bespoke interfaces for users on the fly.

Unlike traditional software, the core of an AI product is its dynamic, often unpredictable output. Static wireframes, even with placeholder text, are mere 'gargoyle rain spouts'—decoration that fails to represent the actual system. You can't validate an AI idea without building and testing the real, content-generating thing.

Designers need to get into code faster not just for prototyping, but because the AI model is an active participant in the user experience. You cannot fully design the user's interaction without directly understanding how this non-human "third party" behaves, responds, and affects the outcome.