Overview
What it is and the problem it addresses.
A reference design for a workflow builder that lets operations teams automate processes with AI steps and approvals, without waiting for engineering.
The problem
Automation tools are either too simple for judgement-based work or too technical for the people who understand the process.
- Visual canvas
- AI agent steps
- Human approvals
- Integrations
- Run monitoring
Main features
What the concept includes.
Canvas
Drag-and-drop triggers, agents, tools, conditions and actions.
AI steps
Classification, extraction and drafting steps with schema-checked output.
Approvals
Thresholds that pause a run until a person decides.
Integrations
Connectors to email, CRM, databases and APIs.
Monitoring
Run history, failures, retries and cost.
Architecture overview
How the pieces fit together.
- 01
Editor
Canvas and versioned workflow definitions.
- 02
Engine
Durable execution with retries and approvals.
- 03
Connectors
Typed integrations and MCP tools.
- 04
Monitoring
Traces, metrics and alerts.
Example workflows
How it is used.
Invoice intake
- 1Invoice arrives by email
- 2AI extracts fields
- 3Rules check amounts
- 4Large invoices wait for approval, then post to the ledger
Target users
- Operations and finance teams
- Customer support
- IT
Potential integrations
- ERP and CRM
- Databases
- Webhooks and MCP
Security considerations
- Role-based edit and run rights
- Secrets kept out of workflows
- Approval gates
- Versioned, auditable changes
Want something like AI Workflow Automation Builder?
Tell us about your goals and constraints. We will outline how a similar system could be built for you.