Triggers Have to Reach the Right Moment
Eyal distinguishes external triggers, which prompt a user from outside, from internal triggers, which fire from within because a habit has formed. For an agent-backed product, the external trigger is often automated — a summary, an alert, a drafted item surfaced at the right moment. Getting that timing right is a workflow problem, not a magic one. Scope the trigger narrowly, build the eval, and measure whether users actually act on it before calling it a feature.
Variable Reward Lives Inside the Eval Loop
The variable reward that keeps users returning is not randomness for its own sake; it is the right answer arriving in a form that feels freshly useful each time. For an agent, that variability has to be earned through good context engineering — retrieving exactly what the task needs rather than dumping everything available. When the output is reliably surprising in the useful direction, the habit forms. When it is surprising in the broken direction, trust collapses and the loop dies.
Investment Compounds Only If Users Change the System
Eyal's most underappreciated insight is that investment — the data, preferences, and history a user loads into a product — makes the next cycle more valuable. In an agent-native product this is structural: every correction a user makes, every eval case a team adds, every boundary a human judgment call creates becomes the investment that tightens the next run. Design explicitly for that accumulation, or the loop stays shallow and churn comes back anyway.
