Skip to Content
🎉 Scout Docs 2.0 is here!
AgentsAgent Delegation

Agent Delegation

Build sophisticated multi-agent systems where specialized agents collaborate to accomplish complex tasks.

What is Agent Delegation?

Agent delegation allows one agent to delegate tasks to other specialized agents. Instead of building one agent that does everything, you can build a team of focused specialists that work together.

This enables patterns like:

  • Specialization — Each agent excels at a specific type of task
  • Quality control — Multiple agents review and validate outputs
  • Complex workflows — Agents coordinate multi-step processes
  • Expertise layering — Combine research, analysis and writing agents

Why Delegate?

Specialized Agents Perform Better

A general-purpose agent making decisions across many domains is less effective than specialized agents:

❌ One agent does everything:
   "Analyze this sales call, update the CRM, draft follow-up, 
    research competitor alternatives and create a deal strategy"

✅ Specialized agents collaborate:
   Research Agent → Analyzes call and competitor landscape
   CRM Agent → Updates records with structured data
   Writing Agent → Drafts personalized follow-up
   Sales Strategist Agent → Creates deal strategy

Panel of Judges Pattern

Use multiple agents to review and validate important decisions:

  1. Agent A makes an initial decision
  2. Agents B, C and D review from different perspectives
  3. Final action based on consensus or majority vote

This dramatically improves accuracy for high-stakes decisions.

Scalable Architecture

As your needs grow, add more specialized agents instead of overloading one agent:

Month 1: 1 research agent
Month 3: +1 writing agent, +1 analyst agent
Month 6: +1 reviewer agent, +1 CRM specialist agent

How Delegation Works

Agent-to-Agent Communication

When an agent delegates to another agent:

  1. Task handoff — Agent A identifies a subtask for Agent B
  2. Context passing — Relevant information is passed to Agent B
  3. Execution — Agent B handles its specialized work
  4. Return results — Agent B sends results back to Agent A
  5. Integration — Agent A incorporates Agent B’s work

Example Flow

User: "Research Acme Corp and prepare for my sales call tomorrow"

Main Agent:
├─ Delegates to Research Agent: "Research Acme Corp company info, 
│  recent news and key decision makers"
│  └─ Returns: Company profile, news summary, org chart
│
├─ Delegates to CRM Agent: "Check our interaction history with Acme"
│  └─ Returns: Past calls, emails, deal stage, notes
│
├─ Delegates to Writing Agent: "Draft a call prep brief using 
│  research and CRM data"
│  └─ Returns: Structured briefing document
│
└─ Returns to user: Complete call preparation package

Common Delegation Patterns

Research → Analysis → Output

A pipeline pattern where each agent handles one phase:

Research Agent → Gathers raw information
    ↓
Analyst Agent → Processes and synthesizes
    ↓
Writer Agent → Creates final deliverable

Use when: Tasks have clear phases that each need different skills.

Primary + Specialists

One agent acts as coordinator, delegating to specialists:

Coordinator Agent
├─ Delegates to Salesforce Agent for CRM tasks
├─ Delegates to Gmail Agent for email tasks
├─ Delegates to Calendar Agent for scheduling
└─ Delegates to Research Agent for web research

Use when: You have a primary workflow that occasionally needs specialized help.

Review Chain

Output passes through multiple review agents:

Draft Agent → Creates initial content
    ↓
Fact-Check Agent → Verifies accuracy
    ↓
Brand Voice Agent → Ensures tone consistency
    ↓
Final Output

Use when: Quality is critical and review from multiple angles adds value.

Parallel Processing

Multiple agents work simultaneously and combine results:

           ┌→ Pricing Agent → price analysis
Input ─────┼→ Tech Agent → technical assessment
           └→ Support Agent → support requirements
                      ↓
              Combine results → Final recommendation

Use when: Multiple independent analyses can run in parallel.

Setting Up Delegation

Part 1: Enabling Your Agent to Delegate to Other Agents

  1. Navigate to your agent in Scout Studio
  2. Click on the Add tool button
  3. Scroll the panel to find the delegate to agent tool
  4. Select the tool to add it to your agent
  5. Once added, your agent can now delegate tasks to other agents
  6. Update your agent’s instructions to let it know it has delegation capabilities

Part 2: Enabling Your Agent to Receive Delegations

For an agent to receive delegated tasks from other agents, it needs to be discoverable and accessible within your workspace. Here’s how to make an agent “delegatable”:

  1. Share within workspace — Ensure your agent is visible to team members (not private-only)
  2. Clear specialization — Give your agent a descriptive name and description so other agents know when to delegate to it
  3. Well-defined instructions — Your agent’s instructions should clearly state its specialty so delegating agents understand its capabilities

When another agent uses the “delegate to agent” tool, it can see and select from available agents in your workspace. The more clearly defined your agent’s purpose, the more likely it will receive appropriate delegations.

Delegation Triggers

Agents can delegate based on:

  • Explicit instruction — “Ask the Research Agent to look into this”
  • Automatic detection — Agent recognizes a task type and delegates accordingly
  • Rule-based — When certain conditions are met, delegate to specific agent

Passing Context

When delegating, specify what context to share:

Delegation:
  to: research-agent
  task: "Research the company mentioned in the user's question"
  context:
    - user_question
    - previous_conversation (last 5 messages)
    - known_entities

Handling Results

Define how the delegating agent should use results:

On Result:
  action: incorporate
  format: "The Research Agent found: {result}. Based on this, ..."

Best Practices

Define Clear Specializations

Each agent should have a narrow, well-defined scope:

  • ✅ “I analyze financial statements and extract key metrics”
  • ❌ “I help with finance stuff”

Establish Communication Protocols

Standardize how agents communicate:

  • What format should results be in?
  • What context should always be passed?
  • How should errors be handled?

Avoid Delegation Loops

Ensure delegation flows in one direction:

✅ Agent A → Agent B → Agent C
❌ Agent A → Agent B → Agent A (loop)

Add cycle detection if agents can delegate to each other.

Monitor Performance

Track:

  • Which delegations are most common
  • Success rates for different agent combinations
  • Time added by delegation vs. value gained

Fallback Behavior

Plan for when delegated agents fail:

  • Timeout handling
  • Alternative agents to try
  • Graceful degradation to original agent

Limitations

  • Each delegation adds latency (agent must wait for response)
  • Complex delegation chains can be harder to debug
  • Not all tasks benefit from delegation. Some are faster done directly.

Next Steps

Built with ❤️ by Scout OS

Last updated on
Planning ToolsAgent Blocks