What a workflow is
A workflow is the unit of work in Tray.ai. Each workflow has:
- A trigger — an event or schedule that starts it. A new Salesforce lead, a Slack message, a daily 6am run.
- Steps — discrete actions along the way. Call a connector. Transform data. Branch on a condition. Wait for human approval. Invoke an AI agent.
- Logic — branching, looping, error handling, retries.
- Observability — every step logged. Every run (task) traceable.
Workflows run in Tray.ai’s managed runtime. You build them visually; they execute at scale.
What makes a workflow modern
The old pattern was “glue code” — scripts that call APIs and move data. It works until it doesn’t: debugging, observability, version control, team collaboration, reliability — each becomes a separate pain.
Modern workflows treat each of those concerns as first-class:
- Visual builder for readability and team collaboration.
- Built-in retry, error handling, alerting.
- Observable execution — know what ran, how long, what failed.
- Versioning, rollback, environment promotion (dev → staging → prod).
- Integrations with connectors pre-built.
Workflows at Tray.ai also integrate with AI agents naturally — a workflow can invoke an agent as a step, and an agent can invoke a workflow as a tool through Agent Gateway.
Patterns
Common workflow shapes Tray.ai customers build:
- Lead lifecycle — DocuSign routes 150M tasks/month this way.
- Data integration — IBM collapsed 40 hrs/week into 5 min/day.
- Employee lifecycle — a fast-growing fintech onboards employees in minutes, not days using Tray.ai workflows.
- Order processing — Auctane processes 3B orders/year.
Each is a workflow (or set of workflows) doing the bread-and-butter orchestration job.