Agent-to-Agent Orchestration: Multi-Agent Workflows
Build pipelines where multiple agents collaborate — in parallel, sequentially, or conditionally. Lint, test, review, and deploy with a single API call.
Beyond Single Agents
A single agent is powerful. A team of agents working together is transformative. Orchestrations let you define multi-agent workflows where each agent handles a specific part of the job — then passes results to the next.
How Orchestration Works
An orchestration defines:
- A strategy — how agents are coordinated
- A list of steps — each step runs a specific agent
- Dependencies — which steps depend on others
When you run an orchestration, Computer Agents manages the execution order, passes context between agents, and tracks the status of every step.
Strategies
| Strategy | Description |
|---|---|
| sequential | Steps run one after another, in order |
| parallel | All steps run at the same time |
| conditional | Steps run based on conditions (e.g. "only if linting passes") |
| map_reduce | Split work across agents, then combine results |
Creating an Orchestration
Sequential Pipeline
The most common pattern — a pipeline where each step builds on the previous one:
TypeScript
Python
Running an Orchestration
TypeScript
Python
This kicks off the workflow. Each step runs in sequence, and you can poll for status:
TypeScript
Python
Parallel Execution
When steps are independent, run them in parallel for speed:
TypeScript
Python
All three agents start simultaneously. The orchestration completes when all steps finish.
Conditional Steps
Run steps only when conditions are met:
TypeScript
Python
Deploy only runs if tests pass. The notifier only runs if tests fail.
Map-Reduce
Split work across multiple agents, then combine results:
TypeScript
Python
The same reviewer agent processes each module in parallel, then a summarizer combines everything into one report.
Coordinator Agent
For complex workflows, set a coordinator agent that oversees the orchestration:
TypeScript
Python
The coordinator can make decisions about retries, error handling, and dynamic step adjustments.
Managing Orchestrations
TypeScript
Python
Real-World Examples
CI/CD Pipeline
Strategy: sequential
- Lint code (Haiku — fast and cheap)
- Run tests (Sonnet — thorough)
- Build artifacts (Sonnet)
- Deploy to staging (Sonnet)
- Run smoke tests (Haiku)
Content Production
Strategy: parallel + sequential
- Parallel: Research topic, Find images, Analyze competitors
- Sequential: Write draft, Edit and polish, Generate social posts
Customer Support Triage
Strategy: conditional
- Classify incoming ticket
- If bug: Reproduce and document
- If feature request: Add to backlog and respond
- If question: Search docs and draft reply
What's Next?
- Event-Driven Triggers — Fire orchestrations automatically on events
- Scheduling Tasks — Run agents on a recurring cron schedule
- Custom Environments — Configure the execution environment
Ready to get started?
Try Computer Agents today and experience the future of AI-powered automation.
Get Started