Multi-step automations — with AI in the middle.

A one-step automation moves data; a multi-step one does the work. This lesson chains several apps and drops AI into the flow — to summarize, classify, draft, and route — so the automation makes the small judgments a human used to make. Then you branch on the result and map data cleanly.

The mental model

Real automations chain several steps — and the magic step is AI in the middle.

A one-step automation moves data. A multi-step one does work: it gathers, then an AI step reads and decides (summarize this, categorize that, draft a reply), then it routes the result. That AI-in-the-middle step is what separates dumb plumbing from an assistant that handles the task.

The reframeThe AI step is where judgment goes. Anywhere your old process needed a human to read something and decide, that's a place an AI step can slot in — summarizing, classifying, or drafting.

Step 01 · Pick the AI step

The AI-in-the-middle reads content and produces a decision or output. Pick what this flow's AI step should do for an incoming support email:

Step 02 · Watch the chain run

Multi-step = trigger → gather → AI step → branch → route. Press run to watch it classify and branch:

trigger new email received ✓

gather pulled subject + body ✓

AI step classified → "complaint" ✓

branch complaint → route to #support channel ✓

notify logged + flagged to you ✓

That branch — "if it's a complaint, send to support; otherwise log it and move on" — is where automations start to feel intelligent. Multi-step prompt: "when [trigger], first [step A], then [step B], then [step C]; pass each result into the next; show me the data at each stage."

Step 03 · Map data cleanly, fail loudly

Most breakage happens at the seams between apps — a date in the wrong format, a field that doesn't exist. Be explicit about which field maps to which, and what to do with anything missing.

Multi-step automations fail silently — a step errors and you never notice the leads stopped arriving. Always add a "tell me if anything fails" step that notifies you, and check it for the first week.

Your challenge: build a smart workflow

Chain a trigger into at least three steps.
Add one AI step that summarizes, classifies, or drafts.
Branch the flow based on that AI step's output.
Add a failure notification and run it on real data.

✓

Lesson complete

You can chain a trigger into a multi-step sequence, insert an AI step to summarize/classify/draft/extract, branch on an AI decision, map data cleanly, and catch silent failures.

Up next

Lesson 3 · Production automations — reliability & team rollout

Continue to Lesson 3 →

Open the Playground Back to Automate a Workflow