Building the automation with AgentMaker

AgentMaker also opens inside the automation editor. Instead of dragging steps and filling in fields one by one, you describe what should happen — “when a customer goes 2 days without replying, send a message” — and it builds the flow for you to review.

It is the same AgentMaker that helps you write an agent’s prompt, in a different mode: here it knows the automation steps, plus your tags, lists, inboxes, agents and people — and it builds the flow out of them.

How to open it

In the automation editor, click AgentMaker in the top bar. The panel opens to the left of the canvas.

Two things to know:

• It only shows up in the Editor section. There is no flow to build under Settings or Activity.
• It only shows up for people who can edit automations. Read-only members don’t see the button.

How it works

1. You describe what should happen, in plain language.
2. AgentMaker builds the flow and checks with SquadOS that it is valid before showing you anything.
3. The proposal arrives as a card listing what changes: new steps, changed steps (with each field’s new value), removed steps, and connections added or removed.
4. You click Apply to the flow. The canvas redraws and the automation is left with unsaved changes.
5. Saving and turning it on stay with you. It never saves, never turns anything on, never publishes.

Applying goes into the editor’s history: if you don’t like it, Ctrl+Z puts the flow back the way it was.

It won’t hand you a broken flow

Before showing a proposal, AgentMaker submits the flow to the same check that runs when you save: a trigger with no outgoing step, two triggers that overlap, a variable used in a step where it doesn’t exist yet, a tag that isn’t in your organization. If something doesn’t pass, it fixes the flow and tries again — you only see the result once it is valid.

When it can’t build what you asked for, it says so, instead of handing you a flow that would fail at save time.

If you edit the canvas after the proposal

The proposal remembers the flow it was based on. If you edited the canvas in the meantime and click Apply, the card warns you and does not apply:

You changed the flow after this proposal. To avoid overwriting your edit it was not applied — ask AgentMaker to redo it against the current flow.

That’s deliberate: applying on top would erase your edit with no warning. Just ask it to redo the proposal.

When a tag or a list is missing

If what you asked for depends on a tag or a list that doesn’t exist yet, it doesn’t make one up: a card appears with what’s missing and a Create button. Once created, click Continue — it picks up where it left off, now with the resource available, and builds the flow.

It does not create inboxes or agents. Connecting a channel and creating an agent are different flows; in those cases it explains what to do and where, and you handle it before coming back.

What it doesn’t do

• Save, turn on or publish the automation. A live automation fires on its own and talks to real customers — that click is always yours.
• Create an inbox (that means connecting a channel) or create an agent.
• Touch the automation’s Settings tab.

In all of those it turns into a recommendation: what to do, why, and where to click.

It asks instead of guessing

When something is missing — which inbox to use, how many days to wait, what the message should say, which criterion separates two groups — it asks. Message text goes to your customer, and guessing that would be worse than asking.

If you want the proposal on the first try, say it all at once: “when a conversation comes into the Support inbox, wait 1 hour; if nobody has replied, hand it off to the Triage agent”.

The conversation is kept

Each automation has its own conversation, and it is kept in this browser while you work: reloading the page or coming back later doesn’t lose the context.

To start over, use the restart conversation button at the top of the panel — it asks for confirmation, and clearing the conversation doesn’t change the flow on the canvas. The X closes the panel without erasing anything.

Good practice

• Describe the outcome, not the steps. “I want to alert the team when a customer complains” works better than dictating each node.
• One change at a time. Easier to review the card, and easier to undo.
• Review before saving. The card shows what changes; the canvas shows the whole flow. Save when both look right.
• Test after applying. The Test button runs the flow that is in the editor right now, with simulated sends — see Runs and tests.