多 Agent 协作编排：从单 Agent 到 Agent 团队

为什么需要多 Agent？

单 Agent 面对复杂任务时容易「顾此失彼」——又要调研、又要写稿、还要审核，角色混杂导致输出质量不稳定。多 Agent 的核心思想是分而治之：每个 Agent 专注一个角色，通过编排机制协作完成目标。

方案对比

步骤 1：用 Fast-Agent 实现

from fast_agent import FastAgent

fast = FastAgent("Research-Write-Review")

@fast.agent(
    "researcher",
    instruction="You are a research specialist. Find comprehensive information on the given topic. Return structured notes.",
    servers=["fetch"]
)
@fast.agent(
    "writer",
    instruction="You are a professional writer. Take research notes and write a polished article. Focus on clarity and engagement."
)
@fast.agent(
    "reviewer",
    instruction="You are an editor. Review the article for accuracy, style, and completeness. Suggest specific improvements."
)
@fast.chain(
    name="publish_pipeline",
    sequence=["researcher", "writer", "reviewer"]
)
async def main():
    async with fast.run() as agent:
        result = await agent.publish_pipeline(
            "The impact of AI agents on software development in 2026"
        )
        print(result)

步骤 2：用 CrewAI 实现

from crewai import Agent, Task, Crew, Process

researcher = Agent(
    role="Research Specialist",
    goal="Find comprehensive and accurate information",
    backstory="Expert researcher with 10 years of experience",
    allow_delegation=False
)

writer = Agent(
    role="Professional Writer",
    goal="Transform research into engaging articles",
    backstory="Award-winning tech journalist",
    allow_delegation=False
)

reviewer = Agent(
    role="Senior Editor",
    goal="Ensure article quality and accuracy",
    backstory="Editor at a major tech publication",
    allow_delegation=True
)

research_task = Task(description="Research: {topic}", agent=researcher)
write_task = Task(description="Write article from research", agent=writer)
review_task = Task(description="Review and improve article", agent=reviewer)

crew = Crew(
    agents=[researcher, writer, reviewer],
    tasks=[research_task, write_task, review_task],
    process=Process.sequential
)

result = crew.kickoff(inputs={"topic": "AI agents in 2026"})

步骤 3：Agents as Tools 模式

当任务需要动态路由而非固定流水线时，使用 Agents as Tools：

@fast.agent("researcher", instruction="...", servers=["fetch"])
@fast.agent("writer", instruction="...")
@fast.agent(
    "orchestrator",
    instruction="Route tasks to the right specialist. Research → researcher, Writing → writer. Aggregate results.",
    default=True,
    agents=["researcher", "writer"]
)
async def main():
    async with fast.run() as agent:
        await agent("Write a comparison of React vs Vue in 2026")

注意事项

• Agent 数量不是越多越好——3-5 个 Agent 是最佳范围
• 每个 Agent 的 instruction 要明确边界，避免职责重叠
• 链式编排适合流水线场景，路由编排适合动态分配场景
• 注意 Token 消耗：多 Agent 编排的 Token 成本是单 Agent 的 3-5 倍