Every automation platform demos well. A form submission lands in a spreadsheet, a message posts to a channel, the audience nods. The workflow that goes into production is a different object: in its first week it meets an empty field, a renamed column, a rate limit and a duplicate record. This course spends most of its time on the parts that decide whether an automation survives contact with real data.
You build the same job three times, in Zapier, Make and n8n, so the choice between them stops being a matter of which one you saw first. The differences that matter are practical: how each platform charges for work, how each handles arrays and loops, what happens when a step fails at three in the morning, and whether you can run the thing on your own server. Before any of that you map the process you actually have, including the exceptions people handle by hand and never mention, because automating a broken process only makes it fail faster.
Day one is the foundation. Triggers by webhook, schedule and polling, and why the choice changes the design. Connecting apps with stored credentials and least-privilege access rather than a key pasted into a step. Mapping fields between systems that disagree about dates, names and empty values. Filters, routers and branching so one workflow handles several cases, and looping over a list without starting a runaway execution. You finish the day with an end-to-end workflow running in your own account.
Day two adds the model. You classify inbound enquiries, extract structured fields from invoices and PDFs, and draft replies that a person approves before they are sent. The skill that decides whether any of it works is making a model step return output the next node can rely on: a schema, validation, and a defined path for the times it does not parse. You then add retrieval so a workflow answers from your own documents instead of guessing, build a tool-using agent in n8n, and work out honestly which jobs an agent should not be given.
The last session is reliability and cost, which is where most automations are actually won: retries and error branches, idempotency so a retry does not send the same invoice twice, alerting when something stops, logging that tells you what ran, a human approval step with an audit trail, and an estimate covering both platform operations and model tokens before you switch anything on. You leave with three working automations and a runbook that lets a colleague maintain them. Runs live online, in the classroom in Colombo or privately, priced in LKR for Sri Lanka and USD internationally.
