Designing Multimodal AI Agents

Write your awesome label here.

In this episode of Unprompted, Karen Kaushansky, UX Lead for Gemini Live at Google DeepMind, shares how conversation design has evolved from flowcharts to multimodal LLMs, exploring trust, voice diversity, and what it takes to design AI that feels truly human.

We dive into the evolution of design from deterministic flowcharts to open-ended LLMs, the mechanics of building trust in AI, and the role of multimodality in shaping natural, human-like interactions. Karen reflects on her career arc, the resilience it takes to work at the cutting edge, and how designers today must rethink what it means to guide a “conversation” between people and AI systems.

This episode covers:

How conversational AI is shifting from voice-only to multimodal—incorporating gestures, vision, and emotion
Designing Gemini Live, a real-time companion that can see what you see, translate across languages, and collaborate in the moment
The challenge of creating trust frameworks for LLMs—balancing usefulness, transparency, and safety
Why voice identity and diversity matter when users form ongoing relationships with AI
The shift from deterministic flowcharts to principles, guidelines, and facilitation between humans and AI models
The role of user research and feedback loops in validating hypotheses and iterating products
Lessons from designing communication systems for autonomous vehicles and emerging interfaces

Whether you’re a seasoned practitioner or a student entering the field, this conversation offers a thoughtful look at where conversation design has been—and where it’s heading in an age of generative, multimodal AI.

Karen Kaushansky is a Conversational AI and UX design leader with nearly 30 years of experience across voice, multimodal, and natural user interfaces. Currently at Google DeepMind, she leads UX for Gemini Live, working at the frontier of AI-powered, open-ended conversations across mobile and ambient devices.

Karen has previously shaped products at Microsoft (Cortana), Nuance, Samsung, Magic Leap, Zoox, and more. Her work spans domains from AR to autonomous vehicles. A recognized speaker and voice tech pioneer, Karen blends deep product insight with a long-term vision for how we interact with machines.