Quick Start: 3-Agent Sequential Workflow
Build your first orchestrated workflow in 15 minutes - a research paper summarization pipeline
Goal
Build a research paper summarization pipeline with three specialized agents working in sequence:
Searcher → Downloader → Summarizer
The orchestrator coordinates these agents, passing results from one to the next, and aggregates the final output.
Architecture Overview
The workflow demonstrates core orchestration patterns:
- Task Queue: Orchestrator maintains ordered list of tasks
- Sequential Dispatch: Each agent completes before next starts
- Message Passing: Agents communicate via structured messages
- Result Aggregation: Orchestrator collects and combines outputs
What You'll Learn:
- Create an orchestrator with task dispatch
- Implement agent message handlers
- Connect agents via sequential message passing
- Aggregate results from multiple agents
Implementation Steps
Create Orchestrator Class
The orchestrator manages the task queue and coordinates agent execution:
class Orchestrator:
def __init__(self):
self.tasks = []Create file orchestrator.py with the orchestrator skeleton.
Implement Dispatch Method
Add sequential task dispatch logic:
def dispatch(self, task):
self.tasks.append(task)
return self.execute_sequential()The dispatch method queues tasks and triggers sequential execution.
Create Agent Classes
Define three specialized agents with message handlers:
class SearcherAgent:
def handle(self, message):
return {"papers": ["paper1.pdf", "paper2.pdf"]}Create agents.py with SearcherAgent, DownloaderAgent, and SummarizerAgent classes. Each agent receives a message and returns results.
Connect Agents via Message Passing
Wire agents together in orchestrator:
searcher_result = searcher.handle({"query": "AI orchestration"})
downloader_result = downloader.handle(searcher_result)
summary = summarizer.handle(downloader_result)Each agent's output becomes the next agent's input.
Run Workflow and Verify
Execute the complete pipeline:
python orchestrator.pyExpected output:
Searcher: Found 2 papers
Downloader: Downloaded 2 papers
Summarizer: Generated 2 summariesVerification
Test the workflow with a research query:
python orchestrator.py --query "multi-agent systems"The orchestrator should display:
- Search phase: List of discovered papers
- Download phase: Confirmation of paper retrieval
- Summary phase: Generated summaries for each paper
✅ Success Criteria:
- All three agents execute in sequence
- Each agent receives output from previous agent
- Orchestrator aggregates final results
- No errors in message passing
What You Built
You just built your first orchestrated workflow!
This simple pipeline demonstrates:
- Sequential orchestration: Agents execute in defined order
- Message-based communication: Structured data passing between agents
- Result aggregation: Orchestrator combines outputs
- Extensible pattern: Easy to add more agents or modify workflow
Next Steps
Core Build
Expand to parallel execution, error handling, and dynamic routing
Domain Applications
Apply orchestration to economics research workflows