Skip to content

multi-agent-research.md

Latest commit

 

History

History
152 lines (111 loc) · 3.89 KB

multi-agent-research.md

File metadata and controls

152 lines (111 loc) · 3.89 KB
Raw

🏠 Home › 📘 Guides › 🎯 Use Cases › Multi-Agent Research

← Use Cases ━━●━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Production Code Review →

Use Case: Multi-Agent Research System

Source: Anthropic Engineering Blog - June 2025

Problem

Synthesizing comprehensive research from multiple sources requires:

  • Parallel information gathering
  • Domain specialization
  • Quality synthesis

Solution Architecture

%%{init: {'theme': 'base', 'themeVariables': {'lineColor': '#64748b'}}}%%
flowchart TB
    classDef main fill:#8b5cf6,stroke:#7c3aed,stroke-width:2px,color:#ffffff
    classDef subagent fill:#ec4899,stroke:#db2777,stroke-width:2px,color:#ffffff
    classDef tool fill:#f59e0b,stroke:#d97706,stroke-width:2px,color:#ffffff
    classDef state fill:#10b981,stroke:#059669,stroke-width:2px,color:#ffffff

    USER["🙋‍♀️📥 Research Query"] --> LEAD["🐔 Main Agent"]:::main

    LEAD -->|"🪺 Task"| PLAN["Plan research strategy"]
    PLAN --> SPAWN["Spawn specialized researchers"]

    subgraph PARALLEL["🚂 Parallel Execution"]
        R1["🐦 Academic Researcher"]:::subagent
        R2["🐦 Industry Researcher"]:::subagent
        R3["🐦 News Researcher"]:::subagent
    end

    SPAWN --> R1 & R2 & R3

    R1 --> T1["🔌 Perplexity"]:::tool
    R2 --> T2["🔌 Firecrawl"]:::tool
    R3 --> T3["🔌 WebSearch"]:::tool

    T1 & T2 & T3 --> COLLECT["🐔 Main Agent collects"]:::main
    COLLECT --> SYNTH["🐦 Synthesis Agent"]:::subagent
    SYNTH --> REPORT["✅ Final Report"]:::state
Loading

Patterns Used

Pattern Role
🦑 Orchestrator-Workers 🐔 Main Agent spawns specialized researchers
🚂 Parallel Tool Calling Multiple researchers work simultaneously
🧬 Master-Clone Each researcher has isolated context

Implementation

🐔 Main Agent Prompt

main_agent_prompt = """
You coordinate research by:
1. Breaking query into research domains
2. Spawning domain-specific researchers
3. Collecting and synthesizing results
"""

🐦 Researcher Subagent

researcher_prompt = """
You research {domain} using available tools.
Return structured findings with citations.
"""

Subagent Definition

# .claude/agents/academic-researcher.md
---
name: academic-researcher
description: Researches academic papers and scholarly sources
tools: Read, WebFetch, WebSearch
model: sonnet
---

You are an academic research specialist. Your task is to:
1. Search for relevant academic papers
2. Extract key findings and citations
3. Return structured research notes

Focus on peer-reviewed sources and cite properly.

Why This Works

Aspect Benefit
Specialization Each subagent focuses on one domain
Parallelism Independent searches run concurrently
Isolation Subagents don't pollute each other's context
Synthesis 🐔 Main Agent has full picture for final output

Flow Summary

🙋‍♀️📥 Research Query
    │
    ▼
🐔 Main Agent (Plan strategy)
    │
    ├──► 🐦 Academic Researcher ──► 🔌 Perplexity
    ├──► 🐦 Industry Researcher ──► 🔌 Firecrawl
    └──► 🐦 News Researcher ──► 🔌 WebSearch
            │
            ▼
    🐔 Main Agent (Collect results)
            │
            ▼
    🐦 Synthesis Agent
            │
            ▼
    ✅ Final Report

← Use Cases ━━●━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Production Code Review →