Workflow Agents

Workflow agents are specialized orchestrators that coordinate multiple agents or execute complex, multi-step processes. Unlike regular agents that respond to individual requests, workflow agents use deterministic logic to control how and when other agents are executed, enabling sophisticated educational workflows.

Workflow agents are perfect for complex educational processes that require multiple steps, different types of expertise, or coordination between various educational tasks.

Key Differences from Regular Agents

Regular Agents
Workflow Agents

Respond to individual requests
Use LLM reasoning for all decisions
Work independently with their tools
Provide direct answers to questions
Best for single-task scenarios

Workflow Agent Types

Sequential Workflows

Execute agents one after another, passing results between them for comprehensive educational pipelines.

How Sequential Workflows Work

Agent 1 completes its task and produces output
Agent 2 receives Agent 1’s output as context and builds upon it
Agent 3 receives outputs from both previous agents
Process continues until all agents have executed
Final result combines all agent outputs into a comprehensive deliverable

Perfect for:

Content creation pipelines (research → design → assess)
Curriculum development (outline → content → activities → assessments)
Quality assurance processes (create → review → refine → approve)

Example: Content Creation Pipeline

Research Agent → Curriculum Designer → Assessment Designer

Research Agent gathers information on the topic
Curriculum Designer creates structured learning content
Assessment Designer develops aligned assessments

Scheduled Workflows

Execute agents at specific times or intervals for automated educational tasks.

Schedule Types

Immediate: Execute right away
At Time: Execute at a specific date and time
Cron: Execute on recurring schedules (daily, weekly, etc.)
Interval: Execute every X minutes/hours/days

Perfect for:

Regular curriculum reviews and updates
Automated content quality checks
Periodic assessment analysis
Scheduled report generation

Example: Weekly Curriculum Review

Schedule: Every Monday at 9 AM
Agent: Curriculum Designer
Task: Review and update weekly lesson plans

Loop Workflows

Repeat agent execution until specific conditions are met, perfect for iterative improvement processes.

Termination Conditions

Maximum Iterations: Safety limit (required)
Success Condition: Stop when specific text appears in results
No Changes: Stop when consecutive results are identical
Timeout: Stop after specified time limit
Custom Conditions: Advanced LLM-based evaluation (planned)

Perfect for:

Iterative content improvement
Assessment question refinement
Curriculum optimization
Quality assurance loops

Example: Content Quality Loop

Content Creator → Review → Improve → Check Quality → Repeat until "excellent quality" achieved

Router Workflows

Intelligently route requests to appropriate specialist agents based on content analysis.

How Router Workflows Work

Analyze Request: Use LLM to understand the request type and requirements
Select Agent: Choose the most appropriate specialist agent
Route and Execute: Send the request to the selected agent
Return Results: Provide the specialist’s response to the user

Perfect for:

General help desks that need to route to specialists
Multi-subject educational support
Adaptive learning systems
Intelligent task delegation

Example: Educational Support Router

Student Question → Analyze Subject → Route to Math Tutor OR Science Tutor OR Writing Coach

Parallel Workflows

Execute multiple agents simultaneously for comprehensive analysis or rapid content generation.

Parallel Execution Benefits

Speed: Multiple agents work simultaneously
Comprehensive Coverage: Different perspectives on the same topic
Efficiency: Parallel processing reduces total time
Diverse Outputs: Multiple approaches to the same problem

Perfect for:

Multi-perspective content analysis
Rapid curriculum development
Comprehensive assessment creation
Research from multiple angles

Example: Comprehensive Topic Analysis

Content Creator + Assessment Designer + Curriculum Designer (all working on the same topic simultaneously)

Evaluate/Optimize Workflows

Execute an agent, then have a review agent evaluate and provide feedback, repeating until quality standards are met.

Evaluation Cycle

Execution Agent creates content or completes task
Review Agent evaluates the output and provides feedback
Execution Agent improves based on feedback
Repeat until quality metrics are achieved or cycle limit reached

Perfect for:

Content quality assurance
Assessment validation
Curriculum review and improvement
Iterative educational design

Example: Assessment Quality Assurance

Assessment Designer creates quiz → Assessment Evaluator reviews → Designer improves → Repeat until high quality

Creating Workflow Agents

Basic Setup

Access Workflow Creation

Navigate to Agents → Workflows tab and click “Create Workflow”

Choose Workflow Type

Select the appropriate workflow type based on your educational process needs

Configure Workflow

Set up agents, execution parameters, and termination conditions

Test and Deploy

Test the workflow with sample inputs and deploy for regular use

Sequential Workflow Configuration

Basic Setup
Advanced Options

{
  "type": "sequential",
  "subAgents": [
    { "agentName": "content_creator" },
    { "agentName": "curriculum_designer" },
    { "agentName": "assessment_designer" }
  ],
  "passResults": true,
  "combineResults": true
}

{
  "continueOnError": false,
  "stepTimeoutMs": 300000,
  "stepMessages": [
    "Research the topic thoroughly",
    "Create curriculum based on research",
    "Generate assessments for curriculum"
  ],
  "context": {
    "gradeLevel": "high_school",
    "subject": "biology"
  }
}

Loop Workflow Configuration

Basic Loop
Advanced Loop

{
  "type": "loop",
  "subAgent": { "agentName": "content_creator" },
  "terminationConditions": {
    "maxIterations": 5,
    "successCondition": "content quality is excellent"
  },
  "passResults": true
}

{
  "terminationConditions": {
    "maxIterations": 3,
    "timeoutMs": 600000,
    "noChangesCondition": true,
    "iterationTimeoutMs": 120000
  },
  "delayBetweenIterationsMs": 2000,
  "exponentialBackoff": true,
  "continueOnError": false
}

Scheduled Workflow Configuration

Cron Schedule
One-Time Schedule

{
  "type": "scheduled",
  "targetAgent": { "agentName": "curriculum_designer" },
  "scheduleType": "cron",
  "cronExpression": "0 9 * * 1-5",
  "timezone": "America/New_York",
  "customMessage": "Perform daily curriculum review"
}

{
  "scheduleType": "at_time",
  "executionTime": "2024-12-25T10:00:00Z",
  "maxExecutions": 1,
  "customMessage": "Generate holiday learning activities"
}

Real-World Educational Workflows

Comprehensive Course Development

Sequential Workflow: Complete Course Creation

Agents: Knowledge Engineer → Curriculum Designer → Content Creator → Assessment Designer → Learning TutorProcess:

Knowledge Engineer creates topic hierarchy and skill mappings
Curriculum Designer structures learning progression and modules
Content Creator generates educational materials and resources
Assessment Designer creates aligned assessments and rubrics
Learning Tutor reviews for student accessibility and engagement

Timeline: 2-4 hours for comprehensive course development

Adaptive Content Improvement

Loop Workflow: Iterative Content Refinement

Agent: Content Creator with quality evaluationProcess:

Create initial content version
Evaluate content quality and student appropriateness
Identify areas for improvement
Refine content based on evaluation
Repeat until quality threshold is met (max 5 iterations)

Termination: “Content meets high quality standards” or 5 iterations

Multi-Subject Support System

Router Workflow: Intelligent Student Support

Router Logic: Analyze student question → Route to appropriate subject specialistSpecialists:

Math Tutor (for mathematical concepts and problem-solving)
Science Tutor (for scientific concepts and experiments)
Writing Coach (for language arts and communication)
History Guide (for historical context and analysis)

Benefits: Students get specialized help regardless of subject area

Assessment Quality Assurance

Evaluate/Optimize Workflow: Assessment Validation

Execution Agent: Assessment Designer
Review Agent: Assessment EvaluatorCycle:

Assessment Designer creates quiz/test
Assessment Evaluator reviews for quality, bias, alignment
Assessment Designer improves based on feedback
Repeat until evaluation score > 0.8 or 3 cycles completed

Quality Metrics: Alignment, clarity, appropriate difficulty, bias-free

Scheduled Curriculum Maintenance

Scheduled Workflow: Regular Content Updates

Schedule: Every Sunday at 6 PM
Agent: Curriculum Designer
Task: Review and update weekly lesson plans based on student progress dataBenefits:

Consistent curriculum maintenance
Data-driven improvements
Reduced manual administrative work
Always current educational content

Monitoring and Management

Workflow Status Tracking

Execution Progress
Results Analysis
Schedule Management

Real-time progress updates
Current step and agent status
Estimated completion time
Tool usage and results

Performance Optimization

Execution Time

Monitor workflow duration and optimize agent selection and configuration for efficiency

Resource Usage

Track computational costs and adjust model selection and parameters accordingly

Success Rates

Analyze workflow completion rates and identify common failure points

Quality Metrics

Measure output quality and user satisfaction to improve workflow design

Best Practices

Workflow Design

Start Simple

Begin with 2-3 agents in a workflow before building more complex orchestrations. Test thoroughly at each stage.

Clear Handoffs

Ensure each agent in a sequence receives appropriate context and clear instructions about their role in the larger process.

Error Handling

Plan for agent failures and decide whether workflows should continue or stop when individual agents encounter problems.

Timeout Management

Set appropriate timeouts for individual steps and overall workflows to prevent hanging processes.

Educational Effectiveness

Align with Learning Goals

Ensure workflow outputs directly support specific learning objectives and educational outcomes.

Quality Assurance

Include review and validation steps in workflows that produce student-facing content or assessments.

Accessibility

Consider diverse learning needs and ensure workflow outputs are accessible to all students.

Continuous Improvement

Regularly review workflow performance and update configurations based on educational effectiveness.

Troubleshooting Common Issues

Workflow Hangs or Times Out

Causes: Individual agents taking too long, network issues, complex requestsSolutions:

Reduce individual step timeout limits
Simplify agent prompts and requests
Check network connectivity and service status

Poor Quality Results

Causes: Insufficient context passing, inappropriate agent selection, unclear instructionsSolutions:

Improve context passing between agents
Review agent selection for each step
Clarify step-specific instructions and goals

Workflow Fails Frequently

Causes: Agent incompatibilities, resource limitations, configuration errorsSolutions:

Test individual agents separately
Review error logs for specific failure points
Adjust resource limits and retry policies

Advanced Features

Conditional Logic (Planned)

Future workflow agents will support conditional branching based on agent outputs and educational context.

Human-in-the-Loop (Planned)

Workflows will be able to pause for human review and approval at critical decision points. Workflows will coordinate agents that work with different media types (text, images, audio, video).

Getting Started with Workflows

Identify a Process

Choose an educational process that involves multiple steps or different types of expertise

Map the Workflow

Identify which agents should be involved and in what order or pattern

Start Simple

Create a basic workflow with 2-3 agents to test the concept

Test and Iterate

Run the workflow with sample inputs and refine based on results

Scale and Optimize

Add complexity and optimize performance based on real usage patterns

Begin with sequential workflows for content creation pipelines - they’re the most straightforward to understand and implement effectively.

AI Agents

​Workflow Agents

​Key Differences from Regular Agents

​Workflow Agent Types

​Sequential Workflows

​Scheduled Workflows

​Loop Workflows

​Router Workflows

​Parallel Workflows

​Evaluate/Optimize Workflows

​Creating Workflow Agents

​Basic Setup

​Sequential Workflow Configuration

​Loop Workflow Configuration

​Scheduled Workflow Configuration

​Real-World Educational Workflows

​Comprehensive Course Development

​Adaptive Content Improvement

​Multi-Subject Support System

​Assessment Quality Assurance

​Scheduled Curriculum Maintenance

​Monitoring and Management

​Workflow Status Tracking

​Performance Optimization

Execution Time

Resource Usage

Success Rates

Quality Metrics

​Best Practices

​Workflow Design

​Educational Effectiveness

​Troubleshooting Common Issues

​Advanced Features

​Conditional Logic (Planned)

​Human-in-the-Loop (Planned)

​Multi-Modal Integration (Planned)

​Getting Started with Workflows

Workflow Agents

Key Differences from Regular Agents

Workflow Agent Types

Sequential Workflows

Scheduled Workflows

Loop Workflows

Router Workflows

Parallel Workflows

Evaluate/Optimize Workflows

Creating Workflow Agents

Basic Setup

Sequential Workflow Configuration

Loop Workflow Configuration

Scheduled Workflow Configuration

Real-World Educational Workflows

Comprehensive Course Development

Adaptive Content Improvement

Multi-Subject Support System

Assessment Quality Assurance

Scheduled Curriculum Maintenance

Monitoring and Management

Workflow Status Tracking

Performance Optimization

Best Practices

Workflow Design

Educational Effectiveness

Troubleshooting Common Issues

Advanced Features

Conditional Logic (Planned)

Human-in-the-Loop (Planned)

Multi-Modal Integration (Planned)

Getting Started with Workflows