Skip to content
The Observatory

Building & Agents · 2 min read

The Habit Loop Is Now an Agent Loop

Nir Eyal's Hook Model — trigger, action, variable reward, investment — was written for human psychology, but its structure maps almost perfectly onto how users build trust with agentic products, and founders who see that correspondence early will design stickier, more reliable systems than those who treat retention as a marketing problem.

Featuring · studied & reframed for AI-native builders“Hooked” — Nir Eyal

Triggers Still Fire, But Agents Answer Them

In Hooked, external triggers bring users in and internal triggers keep them coming back. In an agentic product, the trigger is often a recurring job — a report due, a decision pending, a queue filling up. The agent is the thing that answers. Design the trigger surface carefully: what signal tells your agent to act, and does the user feel that signal as relief or intrusion? If the agent fires on noise instead of real need, it trains users to ignore it, and ignored agents are dead agents.

Variable Reward Lives in the Output, Not the Interface

Eyal's variable reward is the unpredictable value that keeps users returning — the useful result that isn't guaranteed but arrives often enough to form a habit. For agentic software, this maps to output quality variance. An agent that sometimes surfaces a genuinely surprising insight, a shortcut, a catch the user would have missed — that is the variable reward. You cannot manufacture it with animation or notification badges. You earn it through evals that measure real usefulness, not task completion. Build the harness before you ship the agent; let the cases where output delights or fails drive what you optimize.

Investment Locks In Through Accumulated Context

The investment phase in Hooked is where users put something of themselves into the product — preferences, history, data — making it more valuable for the next cycle. Agents make this structural. Every task a user hands off trains the system's context: their vocabulary, their standards, their edge cases. That accumulation is the moat. It is not network effects in the old sense; it is personalized context depth that a new competitor cannot replicate on day one. Context engineering — giving the model exactly what it needs, curated rather than dumped — is how you make that investment compound rather than decay.

Judgment Is the Ethical Anchor Eyal Couldn't Fully Solve

Eyal spends the final chapters of Hooked wrestling with the ethics of designing for compulsion, landing on the idea that builders should only make products they would use themselves and that genuinely serve users. That framing was always a little thin, but it becomes load-bearing in agentic software. When an agent acts autonomously, the ethical weight shifts from the user's choice to the founder's design. Humans must set the goals, own the accountability, and define the boundaries of what the agent is permitted to do. Habit formation without that human layer isn't stickiness — it's dependency on a system no one is watching.

Try the Desktop OS

Explore Atlanta’s startup scene, live.

Everything you just read lives inside our Desktop OS — an interactive workspace that opens right in your browser. Its Atlanta desk maps the city’s startup scene: who’s building, what’s happening, and where. Open it and take a look.

Open the Desktop OS →