How Human Habits Shape AI Assistant Behavior

Denis Mitrofanov is a Senior Systems Designer at Lightspeed. For the past two years, he has been designing AI assistants for cross-platform, high-traffic products: behavior, interface, content, and components. Before that, he worked on Yandex Browser, led design at RUSSPASS, and started his own agency. We build an AI assistant into a product and expect users to ask the questions it can answer. Instead, people seem to ask about anything and behave unpredictably. They hand over large, structured tasks, type a single short, vague question, want control over decisions yet expect a human conversation. As a result, product teams don't know what to rely on when designing the assistant. In fact, this behavior is driven by habits and expectations formed by communicating with people and other services. ChatGPT taught people they can ask anything. Letters taught them to lay out a task in full; search engines, to get by with a couple of words. Bank bots taught them to pick from ready-made options. And in a live conversation, you can interrupt and jump to another topic. People bring all of this with them before they send their first message. In the talk: — The evolution of communication patterns: between people, and between people and machines. — Where assistants break expectations, and how to fix it. — Designing an assistant's interface around people's habits.