---
title: "Agent Design Principles"
type: concept
tags: [multi-agent, agent-design, the-agency]
sources: [contributing_zh-cn, contributing-to-the-agency]
last_updated: 2026-04-24
---

# Agent Design Principles

The five core design principles for building high-quality AI agents in The Agency framework.

## Five Principles

### 1. 🎭 Distinct Personality
- Give each agent a unique tone and persona
- Avoid generic "I am a helpful assistant" — be specific and memorable
- Example: "I will find 3-5 problems by default and ask for visual evidence" (Evidence Collector)

### 2. 📋 Clear Deliverables
- Provide actionable code examples
- Include templates and frameworks
- Show real output, not vague descriptions

### 3. ✅ Success Metrics
- Include specific, quantifiable metrics
- Example: "Page load time under 3 seconds on 3G network"
- Example: "10,000+ combined karma points across accounts"

### 4. 🔄 Verified Workflows
- Step-by-step processes that are clear
- Validated in real-world scenarios
- Reject pure theory without testing

### 5. 💡 Learning & Memory
- Agents identify which patterns to recognize
- How they iterate and optimize over time
- What they remember between sessions

## Anti-Patterns (Avoid)
- ❌ Generic "helpful assistant" persona
- ❌ Vague "I will help you..." descriptions
- ❌ No code examples or deliverables
- ❌ Too broad scope (jack of all trades)
- ❌ Untested theoretical solutions

## Connections
- Extends [[Multi-Agent-System-Reliability]] — design-time quality standards complement runtime reliability patterns
- Used by [[Multi-Agent-Team]] — standardized agent creation framework
- Related to [[Agent-Template]] — the structural template implementing these principles