AI-Drafted Workflows: How /ai/draft Works (and Why It Never Auto-Publishes)
Describe an automation in plain language and Workflows generates a draft to review — it never creates or publishes anything live on its own.
You can now describe a workflow in plain language and get a draft back — a starting point built by AI, not a workflow that's already running.
What it does
Describe what you want automated, and the request goes to our AI draft endpoint, which is routed through the platform's own AI service. What comes back is a workflow definition — steps, connections, the shape of the automation — presented for you to look at, edit, and approve. Nothing about this process creates a live, active workflow on its own.
Why it stops at a draft
This is a deliberate choice, not a limitation we plan to lift. A generated automation can touch real systems — sending emails, updating records, calling external APIs — and an AI-generated first draft is exactly the kind of output that should get a human look before it's allowed to run unattended. So the draft sits in front of you like any other workflow you built by hand: you can inspect every step, adjust the logic, wire up credentials, and only then publish it yourself.
Where this fits
Natural-language workflow building has become table stakes across the automation category this year — several other platforms have shipped or announced their own version. What we think is worth calling out is the boundary: ours hands you a draft, full stop, rather than quietly turning a sentence into a running automation. If the draft isn't right, you throw it away or rework it — nothing has happened to your data in the meantime.
This pairs naturally with how we handle human-in-the-loop approval steps elsewhere in Workflows: automation is worth building, but the moment something can act on your behalf, a review step stays in the loop by default.