Skip to main content
Two of the three routes out of New automation save you from an empty canvas. Both produce an ordinary automation you finish and save yourself. Templates in the automations header opens the gallery. There is a search box and a category filter — Greeting & availability, Lead capture, Appointments, Support, Commerce, Follow-up, Integrations. Every template arrives with the copy already written: real messages, real waits, real conditions. What cannot be pre-filled is anything that names a row in your workspace — a tag, a pipeline stage, a form, a Google connection. Those are left empty on purpose; a plausible-looking id would save cleanly and then never match.

​
Requirements on the card

Each card lists what it needs before it can actually run, and whether this workspace has it. There are three kinds, because the fix is different for each: Requirements never block the click. A template you cannot run yet is still worth opening: the editor is where the pickers are. Picking one opens the canvas editor with everything filled in. Nothing is saved until you press Save.

​
Store recipes are ordinary automations

The Messaging tab of a store console offers recipes — order received, order shipped, abandoned cart. Those are not a second engine: each one builds an ordinary automation with a store trigger and a template send, and posts it like the gallery does. It opens on this canvas like anything else, and its deliveries appear in the same run log. The one difference is that a recipe is tied to the store it was built for, so a second connected store does not fire the first store’s messages.

​
Build with AI

Build with AI takes a sentence and returns a draft automation.
1

Describe it

One or two sentences: what starts it, and what should happen. There are suggestions to start from — “Greet first-time WhatsApp customers, then tag them as a new lead”, “Reply outside 9–6 saying we’ll be back in the morning”.
2

Read the draft

You get the trigger it chose, the steps in order, anything you still have to pick, and any notes it left. How it was built expands to show what it did.
3

Open in builder

The draft is handed to the canvas editor. Regenerate tries again from the same description.
Nothing is saved until you press Save in the builder. The draft is editor state, not a row. It is never active, and it goes through exactly the same validation as an automation you built by hand.

​
What it costs

Each generation spends AI credits from the workspace wallet, and the draft card says how many. If the wallet is empty you are told, with a link to top up. A workspace running on its own provider key is billed by that provider instead.

​
What it can and cannot draft

It works from a fixed vocabulary, and anything outside it is left for you:
  • It never invents an id. Tags, segments, pipelines and stages come back blank and are listed under You’ll need to pick these in the builder.
  • Some steps are not available to it: sending an email or writing to a spreadsheet (those are Google actions, whose fields are a contract with a script in your own Google account), starting a flow, running another automation, and sending a WhatsApp template. When the request clearly wanted one, it builds the closest thing it can and says what is missing.
  • Scheduled automations are not offered. A schedule fires for the workspace with no contact attached, so every messaging step under one would be dead configuration.
  • Anything it could not decide, or an assumption it made about a vague request, comes back as a note you can read.
The same thing exists for flows, with its own prompt bar in the flow editor.