Framework

By mapping interactions across two axes — task frequency and task nature — it defines four distinct modes: Explore, Control, Automate, and Assist. Each mode describes a different relationship between the user and the AI, from open-ended strategic collaboration in Explore, to autonomous execution in Automate, to ambient daily insight in Assist. Rather than treating AI as a single capability to be dropped into a product, the framework gives teams a shared language for matching the right AI behavior to the right moment in a user's workflow.

The real benefit is that it turns an abstract question — "where should AI live in our product?" — into a concrete, repeatable design decision. It prescribes not just interaction patterns but the UI components, trust mechanisms, and provenance signals each mode requires, ensuring that AI features are built with the appropriate level of human oversight for the stakes involved. Crucially, the four modes form a continuous cycle: users explore a strategy, validate and commit in Control, let AI execute in Automate, then use Assist to monitor and course-correct — each loop making the system smarter. This cycle framing means teams are designing not just isolated features, but a coherent AI-powered workflow that compounds in value over time.