Designing for AI Agents, Where Behavioural Science Meets Machine Choice

What Changes When an Agent Chooses for Us?

Muz.li's article argues that AI agents add a third decision-maker to every purchase, sitting between the customer and the checkout, and that this middle layer holds authority without human judgement. It draws on a 2026 MIT Media Lab study that tested 17 AI shopping agents across more than 80,000 experiments. Agents shifted their picks when order, rating, price or persuasive messages changed, and they were more easily swayed than people. The piece also leans on UK regulator CMA research on agentic AI and consumers, which warns that agents may not act as faithful servants.

For brand builders, the practical point is that a product page written only for human eyes can fail an agent, because key limits hidden in images, tooltips or PDFs may be missed or misread. The article makes a useful case for clear, complete, honest product information that serves both people and their agents. Strategists, UX designers and product teams planning for agentic commerce should read it before deciding how their catalogues, policies and suitability details should be structured.