Dynamic workflows use the same unified builder as Sequential Workflows, but add hub nodes that can delegate work to other agents at runtime based on context. This lets a single workflow handle complex, branching logic without requiring you to hardcode every possible path.

Sequential vs. Dynamic: What's the Difference?

In a sequential workflow, every agent and every connection is defined upfront. Execution follows the path you drew.

In a dynamic workflow, one or more nodes act as Manager Hubs — they analyze the situation and delegate to the right agents dynamically. The hub coordinates, collects results, and the flow continues from there.

Crucially, these aren't separate workflow types. They live in the same builder. You can have sequential connections before a hub, a hub in the middle, and more sequential steps after it. Mix them however your process requires.

Sequential → [Manager Hub] ⟷ Specialist Agents → Sequential → Output
              (dynamic zone)

When to Use Dynamic Workflows

Scenario	Why Dynamic Works
Routing based on content	Hub decides which specialist to involve
Unknown required steps	Hub determines what needs to happen
Parallel specialist work	Hub can delegate to multiple agents and gather results
Adaptive customer-facing flows	Different requests need different handling
Research across domains	Hub routes to the right subject-matter agents

If your process always follows the same steps in the same order, a Sequential Workflow is simpler. Use dynamic when the right path depends on runtime information.

Making an Agent a Manager Hub

Any task in the workflow builder can become a hub node. When a task's assigned agent is configured to delegate to other agents, it becomes the coordination point for that part of the flow.

Step 1: Configure a Manager Agent

[!TIP] For best results with hub nodes, create a dedicated Manager agent whose sole job is coordination and delegation — not doing domain work itself. Give it a clear system prompt focused on routing, prioritization, and synthesizing results from specialists. A well-scoped Manager agent produces significantly better delegation decisions than a general-purpose agent used for both managing and executing.

In the Agent Builder, configure your Manager agent with:

Clear role definition: "You are a workflow manager. Your job is to delegate tasks to specialist agents and synthesize their results."
Knowledge of available agents: Describe what each specialist can do so the manager can route intelligently
Delegation instructions: Tell it how to decide who to involve and in what order

Step 2: Add the Hub Task to Your Workflow

In the workflow builder, click Add Agent
Select your Manager agent for this task
Give the task a clear name (e.g., "Routing Hub" or "Research Coordinator")

Step 3: Connect Specialist Agents to the Hub

Add tasks for each specialist agent that the hub may delegate to, then connect them to the hub node using delegation edges (drawn from the hub outward to each specialist).

The hub can delegate to one or more specialists depending on what it decides at runtime.

Step 4: Continue with Sequential Steps After the Hub

After the hub has coordinated and collected results, you can connect more sequential tasks downstream:

[Trigger]
    ↓
[Intake Agent]           ← Sequential
    ↓
[Manager Hub] ⟷ [Technical Support Agent]
              ⟷ [Billing Agent]
              ⟷ [Sales Agent]
    ↓
[Response Formatter]     ← Sequential
    ↓
[Output]

The sequential tasks after the hub receive the hub's synthesized output and continue the linear chain as normal.

Sequential Tasks Before a Hub

You can also run sequential agents before the hub — useful for preprocessing or enrichment before the manager decides what to do:

[Trigger]
    ↓
[Data Enrichment Agent]  ← Sequential (always runs)
    ↓
[Classification Agent]   ← Sequential (always runs)
    ↓
[Manager Hub] ⟷ [Specialist A]
              ⟷ [Specialist B]
    ↓
[Output Formatter]       ← Sequential

This lets you normalize or prepare data before the dynamic routing step.

Example: Customer Inquiry Workflow

The Scenario

Incoming customer inquiries can be about billing, technical issues, or sales questions. Each type needs a different specialist, but all inquiries need a final formatted response.

Agents

Intake Agent — Extracts key details from the raw inquiry
Manager Agent (hub) — Analyzes the inquiry and delegates to the right specialist
Technical Support Agent — Handles product/technical questions
Billing Agent — Handles payment and account questions
Sales Agent — Handles pricing and upgrade questions
Response Formatter — Polishes the specialist's answer into a customer-ready reply

Workflow

[Webhook: Incoming Inquiry]
        ↓
[Intake Agent]
  Instructions: "Extract: inquiry type, urgency,
  customer tier, and core question."
  Context: Isolated
        ↓
[Manager Agent] ⟷ [Technical Support Agent]
                ⟷ [Billing Agent]
                ⟷ [Sales Agent]
        ↓
[Response Formatter]
  Instructions: "Format the specialist's answer
  into a professional customer-facing response."
  Context: Full
        ↓
[Output: Customer Reply]

Why This Works

Intake always runs — normalizes input before the manager sees it
Manager delegates to exactly the right specialist based on the inquiry type
Specialists handle their domain without worrying about routing logic
Response Formatter always runs — ensures consistent output regardless of which specialist was involved

Example: Multi-Domain Research Workflow

A more complex example where the hub coordinates parallel research across multiple domains, then a sequential writer synthesizes everything.

[Form: Research Topic]
        ↓
[Research Manager] ⟷ [Academic Sources Agent]
                   ⟷ [Industry Data Agent]
                   ⟷ [News & Trends Agent]
        ↓
[Synthesis Writer]
  Instructions: "Combine all research findings
  into a coherent, well-structured report."
  Context: Full
        ↓
[Editor Agent]
  Context: Full
        ↓
[Output: Research Report]

The manager decides which research agents to involve based on the topic, then the sequential writer and editor always run to produce the final output.

Best Practices

Use a Dedicated Manager Agent

Don't reuse a specialist agent as the hub. A manager agent focused entirely on coordination will make better routing decisions and produce cleaner handoffs to specialists.

Keep Specialist Agents Focused

Each specialist should do one thing well. The manager handles routing complexity; specialists handle domain complexity. Mixing these concerns degrades both.

Give the Manager Context About Specialists

In the manager agent's system prompt or task instructions, describe each specialist's capabilities. The manager can only route intelligently if it knows what each agent can do.

Design Clean Output From the Hub

The hub synthesizes results from specialists before passing to downstream sequential tasks. Give it explicit instructions on how to structure that synthesis — especially if a sequential agent after the hub needs to parse it.

Use Sequential Steps for Guaranteed Execution

Steps that always need to happen (intake, formatting, final output) belong in sequential connections. Reserve dynamic delegation for the parts where routing needs to be adaptive.

Test the Hub With Each Specialist Path

When testing a dynamic workflow:

Test the full pipeline for each possible delegation path
Verify the hub routes correctly for each input type
Confirm sequential steps after the hub receive the synthesized output correctly

Troubleshooting

Hub not delegating to the right agent

Review the manager agent's system prompt — it needs clear descriptions of what each specialist handles
Add explicit routing instructions in the hub task's instructions field
Test the manager agent in isolation with sample inputs to verify its routing logic

Sequential tasks after the hub not receiving output

Confirm the hub task produces structured output that the downstream task can use
Add instructions in the hub's task telling it how to format results for the next step
Check that the connections from the hub to downstream tasks are correct

Specialists running when they shouldn't

Tighten the manager's routing criteria in its system prompt
Verify the delegation edge connections in the canvas

Related Guides

Sequential Workflows — Linear pipelines without dynamic routing
Creating a Workflow — Step-by-step builder walkthrough
Configuring your Agent — Set up agents including Manager agents
Adding Guardrails — Validate outputs at critical steps
Workflow Triggers — Configure how workflows start