Workflow Agent
The orchestrated agent system that builds and validates workflows — from requirement analysis to build planning, flow creation, test case generation, and validation.
The Workflow Agent is an orchestrator that coordinates 3 sub-agents to handle the full lifecycle of creating and validating a workflow. Instead of manually analyzing requirements, building the flow, and writing test cases — the Workflow Agent manages the process through specialized sub-agents.
How it works
When you describe an integration requirement, the Workflow Agent orchestrator:
Analyzes requirements — The Planner Agent asks clarifying questions, identifies required connectors, and produces a structured build plan with test cases.
Builds the workflow — The Workflow Builder creates the flow from the plan, binds triggers, configures steps, and sets up data mappings.
Generates and runs tests — The Test Case Agent creates executable integration tests, runs them against the workflow, and reports results.
Hands back — The orchestrator pauses between steps for user review, then resumes when you confirm.
Diagram needed: Flow diagram showing the Workflow Agent orchestrator delegating: Planner Agent → Workflow Builder → Test Case Agent → Validation. Show handback points between each stage.
GIF needed: Workflow Agent in action — user describes an integration need, the Planner asks clarifying questions, produces a plan, the Builder creates the flow, and the Test Case Agent validates it.
Sub-agents
Planner Agent
What it does: Analyzes integration requirements before any building starts. It asks clarifying questions to understand the full scope, identifies which connectors and actions are needed, determines the trigger type and data flow, and produces a structured build plan with acceptance criteria.
When the orchestrator calls it: First step — before building anything, understand what needs to be built.
Input: A natural language description of the integration requirement (e.g., "When a new order comes in from Shopify, create an invoice in Xero and notify the team on Slack").
Output: A structured build plan containing:
Required connectors and actions
Trigger type and configuration
Step sequence with data mappings between steps
Error handling strategy
Test cases with expected inputs and outputs
How it works in practice:
The Planner doesn't just accept the requirement at face value. It asks clarifying questions:
"Should the flow run for all orders, or only orders above a certain amount?"
"Which Slack channel should receive the notification?"
"What should happen if the Xero invoice creation fails — retry, skip, or alert?"
"Do you need the customer's contact info synced as well, or just the order data?"
Once it has enough context, it produces the plan for the Workflow Builder to execute.
Screenshot needed: Planner Agent conversation showing clarifying questions and the resulting build plan with steps, triggers, and test cases.
Workflow Builder
What it does: Builds the actual workflow from the Planner's build plan. It creates the flow on the canvas, adds and configures each step (connector actions, transformations, control logic), binds the trigger, sets up data mappings between steps, and configures error strategies.
When the orchestrator calls it: After the Planner has produced a build plan and the user has approved it.
Input: The structured build plan from the Planner Agent.
Output: A complete workflow definition — trigger configured, steps added and connected, data mappings set, error strategies applied. The flow appears on the canvas ready for review.
What it configures:
Trigger type and schedule/event binding
Connector steps with the correct actions and credentials
Data propagation — mapping outputs from one step to inputs of the next
Transformation steps for data format conversion
Control logic (conditionals, loops) as specified in the plan
Error strategies per step (halt or continue)
Screenshot needed: A workflow on the canvas that was generated by the Workflow Builder — showing multiple connected steps matching a build plan.
Test Case Agent
What it does: Generates executable integration test cases for the workflow and runs them. It creates test scenarios based on the Planner's acceptance criteria, sets up test input data, executes the workflow in test mode, and validates outputs against expected results.
When the orchestrator calls it: After the Workflow Builder has created the flow — the final validation step before the user deploys.
Input: The completed workflow definition and the test cases from the Planner's build plan.
Output: A test report showing:
Each test case with its input data
Expected vs actual output
Pass/fail status
Step-level execution details for failures
Recommendations for fixes if tests fail
Screenshot needed: Test Case Agent results showing a list of test cases with pass/fail status, input data, expected output, and actual output.
The handback pattern
The Workflow Agent pauses at three points:
After planning: "Here's the build plan — 4 steps, triggered on Shopify webhook, error strategy: retry Xero step 3 times then alert on Slack. 3 test cases defined. Want me to proceed?"
After building: "The workflow is on the canvas. Review the step configuration and data mappings before I run tests."
After testing: "All 3 test cases passed. The flow is ready to deploy. Or I can adjust — what would you like to change?"
Workflow Builder vs. Workflow Agent
There's an important distinction:
What it is
A single agent that builds flows from a plan
An orchestrator managing Planner + Builder + Tester
When to use
You already know exactly what to build — just need it created
You have a requirement but need analysis, planning, and validation
Planning
Expects a plan as input
The Planner sub-agent creates the plan
Testing
Validates against provided test cases
The Test Case Agent generates and runs test cases
Handback
Basic — builds and returns
Full handback at plan, build, and test stages
Use the standalone Workflow Builder (accessible via the AI chat) for quick, well-defined flows. Use the Workflow Agent (Orchestrator) for complex requirements where you need analysis and validation.
When to use the Workflow Agent
"Sync orders from Shopify to Xero" — clear, well-defined
Standalone Workflow Builder is faster
"We need to integrate our e-commerce pipeline" — vague, complex
Workflow Agent → Planner clarifies → Builder creates → Tester validates
You need test coverage for a new integration
Workflow Agent ensures test cases are generated alongside the flow
You're building a workflow template for multiple tenants
Workflow Agent → plan once → build once → test with sample tenant data
An existing workflow needs rework
Planner can analyze what changed and produce an updated plan
Related pages
Creating a Workflow via Chat — Tutorial for the standalone Workflow Builder via AI chat
Writing a Workflow (TypeScript DSL) — Code-based approach for developers
Connector Agent — The equivalent orchestrated system for building connectors
Last updated
Was this helpful?