Appearance
Multi-Agent Parallel Research
Create two specialized agents and run them in parallel on different research tracks within a single workflow.
Internally still called a playbook.
What You'll Use
| Feature | Purpose |
|---|---|
| Multiple Agents | Specialists with different roles and models |
| Parallel Steps | Run branches simultaneously |
| Agent Override | Assign a different agent to a specific step |
Step 1: Create Two Agents
Agent 1: Technical Researcher
Go to Agents and click New Agent.
| Field | Value |
|---|---|
| Name | Technical Researcher |
| Description | Analyzes technical architecture, capabilities, and implementation details |
| Preferred Model | Claude |
| Output Length | Detailed |
Role instructions:
You are a technical researcher. Focus on architecture, technology stack, APIs, performance characteristics, and implementation details. Be thorough and cite specifics.
Agent 2: Business Analyst
Create a second agent:
| Field | Value |
|---|---|
| Name | Business Analyst |
| Description | Analyzes market positioning, business model, and competitive landscape |
| Preferred Model | Gemini 2.5 Flash |
| Output Length | Concise |
Role instructions:
You are a business analyst. Focus on market positioning, pricing strategy, target customers, partnerships, and growth signals. Be concise and highlight actionable insights.
INFO
Using different models for different agents is a cost optimization strategy. Claude excels at nuanced technical analysis, while Gemini Flash is fast and cost-effective for business research that involves processing many sources.
Step 2: Build the Workflow
Go to Workflows and click New Workflow.
| Field | Value |
|---|---|
| Name | Product Launch Research |
| Description | Parallel technical and business analysis of a product or competitor |
| Default Agent | Technical Researcher |
Step 1: Parallel Research (Parallel Step)
Add a Parallel step with two branches:
Branch A — Technical Deep Dive (uses default agent: Technical Researcher)
| Field | Value |
|---|---|
| Instructions | Research the technical architecture and capabilities of the product specified in the input. Cover: technology stack, API design, performance claims, integrations, and any known limitations. |
| Auto-Approve | Yes |
Branch B — Market Analysis (override agent: Business Analyst)
| Field | Value |
|---|---|
| Instructions | Analyze the market positioning and business model of the product specified in the input. Cover: target market, pricing structure, key competitors, recent funding or partnerships, and growth trajectory. |
| Agent | Business Analyst |
| Auto-Approve | Yes |
Step 2: Synthesize Findings (Action Step)
| Field | Value |
|---|---|
| Instructions | Combine the technical and market research into a single briefing document. Start with a one-paragraph executive summary, then include a Technical Assessment section and a Market Assessment section. End with three strategic recommendations. Save as a report artifact. |
| Auto-Approve | Yes |
TIP
The synthesis step uses the default agent (Technical Researcher / Claude) because combining and structuring information benefits from a stronger reasoning model.
Step 3: Run It
- Open the workflow and click Run.
- Enter the product or company name when prompted.
- Watch the Runs tab — both branches of Step 1 execute simultaneously.
- When both branches complete, Step 2 runs and produces the final report.
In the run details, you can see which agent handled each branch and how much each one cost.
When to Use Multiple Agents
| Scenario | Why it helps |
|---|---|
| Different expertise needed | A legal reviewer and a technical writer produce better results than one generalist |
| Cost optimization | Use cheaper models for data gathering, expensive models for synthesis |
| Speed | Parallel branches run simultaneously, cutting total time |
| Different output styles | A concise analyst and a detailed researcher produce complementary outputs |
What to Try Next
- Agents — Risk tolerance, model selection, and role instructions
- Steps — Parallel, decision, and approval step types
- Configuring Agents — One job per agent, model matching
