Multimodal Collaborative Agents for Next-Gen Commerce — Nidhi Kaushik Vyas, Google DeepMind

A Google DeepMind talk proposes that shopping agents choose their next question by how much uncertainty it resolves. In the publisher's living-room example, asking for the room's width comes before recommending furniture because that answer determines what can fit. The proposed workflow combines conversation and reference images, distinguishes firm requirements from inferred preferences, and tracks confidence in each. For subjective choices, it can present a visual board and use the shopper's reactions to refine its understanding. The design gives the agent responsibility for clarifying an incomplete request before searching for products. Talk description.