Agent Design

Understanding how Crowdi agents think and behave is key to interpreting simulation results.

Persona System

Each agent is instantiated with a unique persona that defines:

• Demographics - Age, location, occupation
• Technical Literacy - Beginner, intermediate, or expert
• Goals - What they're trying to accomplish
• Patience Level - How quickly they give up on friction
• Device Preference - Desktop, mobile, or tablet

Cognitive Architecture

Agents follow a continuous decision-making loop:

1. Observe - Analyze the current page state, visible elements, and available actions
2. Plan - Determine the best next action based on their goal and past experience
3. Act - Execute the chosen action (click, type, scroll, navigate)
4. Reflect - Evaluate if the action moved them closer to their goal

Decision Making

Agents use multiple factors to make decisions:

• Visual prominence of UI elements
• Semantic meaning of text and labels
• Past experiences from previous simulations
• Common web interaction patterns
• Their persona's technical literacy level

Failure Modes

Agents can fail in realistic ways:

• Getting stuck in loops
• Abandoning tasks due to friction
• Misunderstanding unclear UI
• Making incorrect assumptions

These failures are valuable - they highlight real UX issues!