Track 5

AI Automation

Put Claude to work while you sleep

Wire Claude into no-code automation tools like Make, n8n, and Zapier. Trigger workflows from email, forms, and schedules, let Claude do the thinking, then write results back to your apps — no babysitting required.

WhereA no-code automation platform — Make, n8n, or Zapier — plus a Claude (Anthropic) API key. Free tiers on all three are plenty to build your first working automation.
Automation · ScenarioLive
New form submissionTrigger · runs automatically
Claude: classify & summarizeReads the message, returns clean JSON
Add row to Google SheetLogs the structured result
Send Slack alertPings the team if it's urgent

The big picture

What AI automation can do

Chat and Cowork are hands-on — you're in the loop. Automation is hands-off: you build a workflow once, and it runs itself whenever something happens. Claude becomes the 'brain' step in the middle.

Trigger on anything

Start a workflow from a new email, form submission, file upload, calendar event, or webhook — no manual kickoff.

Claude as the brain

Drop a Claude step into the middle to classify, summarize, extract, draft, or decide — then act on its answer.

Read & write your apps

Pull data from one app, transform it with Claude, and push the result to Sheets, a CRM, Slack, or a database.

Run on a schedule

Have a workflow fire every morning, every hour, or on a cron — daily digests, reports, and reminders run themselves.

Branch on the result

Route differently based on what Claude returns — escalate urgent items, file the rest, all automatically.

Run unattended, safely

Add filters, error handling, and approval steps so automations fail loudly instead of silently doing the wrong thing.

Your roadmap

From first scenario to production pipelines

Work through the three levels in order. You'll build a real, running automation in the Beginner level, then make it robust and powerful.

Build something that runs itself

  1. 1

    Pick your platform

    Start with Make, n8n, or Zapier — all have free tiers and a visual canvas. Make and n8n give you the most control; Zapier is the simplest. Any of them works for this track.
  2. 2

    Get a Claude API key

    Create a key at console.anthropic.com. Automation calls the Claude API (pay-per-use), which is separate from your claude.ai chat subscription. Add a few dollars of credit to start.
  3. 3

    Add a trigger

    Every workflow starts with a trigger — the event that kicks it off. Pick something simple first, like “new email in a Gmail label” or “new row in a Google Sheet.”
  4. 4

    Add a Claude step

    Add an Anthropic / Claude module (or an HTTP call to the Messages API). Feed it data from the trigger and write a clear instruction: “Summarize this email in one sentence and label it urgent or normal.”
  5. 5

    Do something with the answer

    Add a final action that uses Claude's output — append it to a sheet, post it to Slack, or send a reply. Then run once to test the whole chain end to end.
  6. 6

    Turn it on

    Flip the scenario to active. It now runs on its own every time the trigger fires. Congratulations — that's a working AI automation.

Start absurdly small

Your first automation should do one boring, useful thing — like logging form submissions to a sheet with a one-line summary. Get that running before you reach for anything clever. Momentum beats ambition here.

Keep it handy

Quick reference

Common patterns

PatternShapeGreat for
Classify & routeTrigger → Claude → routerTriaging emails, tickets, and leads automatically.
Extract to dataTrigger → Claude (JSON) → Sheet/DBTurning messy text into clean, structured rows.
Scheduled digestCron → gather → Claude → emailDaily summaries, reports, and standups.
Draft & approveTrigger → Claude → human → sendReplies and outreach that need a sign-off.
Enrich a recordTrigger → Claude → update CRMFilling in missing context on new contacts.

Demand JSON

Structured output is the difference between a demo and a pipeline.

Let Claude decide, the flow route

Keep logic in the workflow; keep judgment in the prompt.

Test with 'run once'

Walk the whole chain on real data before going live.

Alert on failure

Silent automations rot. Make errors page you.

Vault your keys

API keys belong in connections/secrets, never inline.

Small, chained scenarios

Modular pipelines beat one giant unmaintainable flow.

Common questions

Good to know