Workflows are where everything comes together. They transform AI understanding into reliable, repeatable automation by combining context awareness , secure tool execution , and intelligent orchestration into structured processes your team can trust.
What Makes Kubiya Workflows Different AI-Generated, Human-Governed Unlike traditional automation that requires manual scripting, Kubiya workflows start with natural language:
You say : “Deploy the payment service to staging and run tests” Kubiya creates : A workflow that uses your existing tools (kubectl, helm, etc.) and follows your deployment patterns, but executes reliably every time.Deterministic Execution Once generated, workflows execute the same way every time:
Same input → Same output : No variation between executionsPredictable behavior : Every step follows defined logicReproducible results : Easy to debug and troubleshootAudit trail : Complete record of every action taken
Context-Aware Intelligence AI generates workflows that understand your specific environment by reading your existing configurations, deployment patterns, and infrastructure setup.
Workflow Creation Methods 1. Natural Language Generation Start with a conversational description:
Examples of effective prompts:
“Scale down development environments after 6 PM to save costs” “When CPU usage exceeds 80% for 5 minutes, automatically scale up” “Create a runbook for investigating database connection issues” “Set up automated testing for every pull request”
2. Visual Workflow Designer Build complex workflows with drag-and-drop components:
The visual designer includes:
Flow Control Conditions, loops, parallel execution, error handling
Tool Integration Direct access to all your connected systems and tools
Data Transformation JSON processing, templating, data validation, formatting
Human Approval Manual approval steps for sensitive operations
3. Code-First Workflows Define workflows as code for version control and automation:
from kubiya import Workflow, Step # Python DSL for complex logic @workflow def incident_response ( severity : str , service : str ): """Automated incident response workflow""" # Gather initial information logs = Step( "collect-logs" ).tool( "log-aggregator" ).inputs( service = service, timeframe = "30m" ) metrics = Step( "get-metrics" ).tool( "monitoring-client" ).inputs( service = service, metrics = [ "cpu" , "memory" , "error_rate" ] ) # Parallel information gathering info_gathering = Parallel([logs, metrics]) # Analysis step that uses gathered data analysis = Step( "analyze-issue" ).tool( "ai-analyzer" ).inputs( logs = logs.outputs.log_content, metrics = metrics.outputs.metric_data, severity = severity ).depends_on(info_gathering) # Conditional remediation based on analysis if analysis.outputs.recommended_action == "scale" : remediation = Step( "auto-scale" ).tool( "scaler" ).inputs( service = service, scale_factor = analysis.outputs.scale_factor ) elif analysis.outputs.recommended_action == "restart" : remediation = Step( "rolling-restart" ).tool( "restarter" ).inputs( service = service, strategy = "rolling" ) else : remediation = Step( "manual-intervention" ).tool( "pager" ).inputs( message = f "Manual intervention required: { analysis.outputs.details } " ) return Workflow([info_gathering, analysis, remediation]) 4. Template-Based Creation Start with proven patterns and customize:
DevOps Templates Deployment pipelines, rollback procedures, health checks
SRE Templates Incident response, capacity planning, chaos testing
Security Templates Vulnerability scanning, compliance checks, access reviews
Cost Optimization Resource cleanup, rightsizing, budget alerts
Advanced Workflow Features Error Handling & Recovery Workflows include comprehensive error handling:
steps : - name : deploy-application tool : kubernetes-deployer retry : max_attempts : 3 backoff : exponential on_failure : - name : collect-failure-logs tool : log-collector - name : rollback-deployment tool : kubernetes-rollback - name : notify-team tool : slack-notifier inputs : channel : "#incidents" message : "Deployment failed, rolled back automatically" Conditional Logic & Branching Smart workflows adapt based on conditions:
steps : - name : check-environment-load tool : monitoring-check - name : choose-deployment-strategy condition_branches : - condition : ${check-environment-load.cpu_usage} > 80 steps : - name : gradual-rollout tool : canary-deployer inputs : traffic_split : [ 5% , 20% , 50% , 100% ] - condition : ${check-environment-load.cpu_usage} < 40 steps : - name : fast-deployment tool : blue-green-deployer - default : true steps : - name : standard-rollout tool : rolling-deployer Parallel Execution Optimize performance with concurrent operations:
# Run multiple health checks in parallel parallel_health_checks : - name : database-health tool : postgres-checker - name : cache-health tool : redis-checker - name : api-health tool : http-checker inputs : endpoints : [ "https://api.example.com/health" ] # Wait for all parallel tasks before proceeding - name : aggregate-results tool : health-aggregator depends_on : [ database-health , cache-health , api-health ] Human Approval Gates Include manual checkpoints for sensitive operations:
- name : request-production-approval tool : approval-gate inputs : approvers : [ "@platform-team" , "@security-team" ] timeout : 2h approval_message : | Production deployment requested: - Service: ${SERVICE_NAME} - Version: ${BUILD_VERSION} - Impact: ${ESTIMATED_IMPACT} - Rollback plan: Available - name : deploy-to-production tool : production-deployer condition : ${request-production-approval.approved} == true depends_on : [ request-production-approval ] Workflow Execution & Monitoring Real-Time Execution Tracking Monitor workflows as they run:
Track key metrics:
Execution time per step and overall workflowSuccess rate and failure patternsResource usage during executionCost per execution across different environments
Alerting & Notifications Stay informed about workflow execution:
notifications : on_start : - slack : "#deployments" message : "🚀 Starting ${WORKFLOW_NAME}" on_success : - slack : "#deployments" message : "✅ ${WORKFLOW_NAME} completed successfully" - email : "platform-team@company.com" on_failure : - slack : "#incidents" message : "❌ ${WORKFLOW_NAME} failed at step ${FAILED_STEP}" - pagerduty : service : "platform-automation" severity : "warning" on_human_approval_needed : - slack : "@channel in #approvals" message : "🔒 Approval required for ${WORKFLOW_NAME}" Workflow Lifecycle Management Version Control Integration # .kubiya/workflows/payment-deployment.yaml apiVersion : kubiya.ai/v1 kind : Workflow metadata : name : payment-service-deployment version : "2.1.0" labels : team : payments criticality : high environment : production Testing & Validation Test workflows before production use:
# Workflow testing framework import pytest from kubiya.testing import WorkflowTest class TestPaymentDeployment ( WorkflowTest ): def setUp ( self ): self .workflow = self .load_workflow( 'payment-deployment' ) self .mock_environment( 'staging' ) def test_successful_deployment ( self ): # Mock successful responses from all tools self .mock_tool_success( 'kubernetes-deployer' ) self .mock_tool_success( 'test-runner' ) result = self .workflow.execute() assert result.status == 'completed' assert len (result.steps) == 4 assert result.steps[ - 1 ].name == 'production-deployment' def test_rollback_on_test_failure ( self ): # Mock test failure self .mock_tool_failure( 'test-runner' , error = 'Integration tests failed' ) result = self .workflow.execute() assert result.status == 'failed' assert 'rollback' in [step.name for step in result.executed_steps] Best Practices Workflow Design Principles
Idempotent Operations Design steps that can be safely re-run without side effects
Fail Fast Check prerequisites early to avoid wasting time and resources
Clear Dependencies Make step dependencies explicit for reliable execution order
Comprehensive Logging Log all decisions and actions for debugging and audit
Security Considerations security : # Least privilege - only necessary permissions permissions : - resource : "deployments" namespace : "payment-service" actions : [ "get" , "update" , "patch" ] # Sensitive data handling secrets : - name : database_password source : vault://production/database/password inject_as : environment_variable # Approval requirements for sensitive operations approvals : - condition : environment == "production" required_approvers : 2 allowed_approvers : [ "@platform-team" , "@security-team" ] Pro Tip : Start with read-only workflows to build confidence, then gradually add write operations as your team becomes comfortable with the automation patterns.
What’s Next? With workflows providing structured automation, explore real-world use cases that show how teams apply these concepts to solve common infrastructure and operations challenges.
