Role-Based Agents

Introduction

Role-based agents embody specific personas — Researcher, Coder, QA, Designer. Each has a tailored system prompt and tool set, narrowing scope and improving quality.

Beginner analogy: a company hires roles, not 'generalists' — same logic for agents.

Understanding the topic

Core concepts:

• Each role = system prompt + tool set + memory.
• Smaller scope → fewer hallucinations.
• Easier evals per role.
• Common roles: planner, executor, critic, summariser, fact-checker.
• Roles can be swapped or upgraded independently.

Syntax reference

Visual workflow / architecture:

┌──────────┐  ┌──────────┐  ┌──────────┐
│ Planner  │  │ Coder    │  │  QA      │
└──────────┘  └──────────┘  └──────────┘
   role:        role:         role:
   plan tasks   write diff    run tests

Real-world use

MetaGPT explicitly models PM/Engineer/QA; CrewAI is built around roles.

Best practices

• Name roles after real-world jobs.
• Give each role its own minimal tool set.

Common mistakes

• Giving every role every tool — defeats specialisation.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Why specialise agents by role?
• Q2. Three common roles in an engineering crew.
• Q3. Scenario: design roles for a marketing agent team.