← All modules
05Intermediate

AI Automation & Tools

Connect n8n and Make to LLMs, and delete the work nobody should be doing by hand.

Every organisation has a person who spends Monday morning copying rows between two tools. That work is not a staffing problem, it is an unbuilt workflow — and until recently automating it meant a developer and a budget. It no longer does. This module treats n8n and Make as what they are: a programming environment with boxes instead of syntax, which removes the typing but none of the thinking. You still have to handle an API that returns an error at 2am, an LLM step that costs money on every run, and a duplicate that fires your workflow twice. That is where most automations die, and that is where we spend the time. The strong opinion here: do not put a model in the loop until you have proved a deterministic step cannot do the job. Models are for the fuzzy middle — classifying a message, extracting a field from unstructured text, drafting a reply. Everything around them should be boring, testable and cheap. You will automate your own real work, not a demo, because a workflow that survives your own inbox is the only proof that counts.

LevelIntermediate
Duration6 weeks
Sessions5
Price1,200 DH

Prerequisites

  • No programming required — but comfort with logic, conditions and data
  • A real repetitive task you want to automate
  • Access to the tools you use daily (email, spreadsheet, CRM)
  • Basic understanding of what an API is

What you can do afterwards

  • Build multi-step workflows in n8n and know when Make is the better fit
  • Connect apps and APIs, including ones with no ready-made integration
  • Place an LLM step only where a deterministic step cannot do the job
  • Handle errors, retries, duplicates and rate limits like a professional
  • Keep a workflow’s running cost predictable and defensible
  • Hand an automation to a colleague who did not build it

Sessions

Mapping the work before automating it

The most expensive automation is the one that perfectly encodes a broken process. Before touching a tool you map the task as it actually happens — including the exceptions everyone handles from memory and nobody wrote down. Half the tasks people bring here should be deleted, not automated.

Covered

Writing down a process as it is, not as it is describedFinding the exceptions nobody documentedDeciding what to delete before what to automateTriggers: schedule, webhook, or eventData in, data out: defining the contractEstimating the hours a workflow actually returns

n8n as a programming environment

Boxes instead of syntax, but the same discipline. You will build real multi-step flows: branching, looping over items, transforming data between shapes, and the code node you reach for when the visual path gets absurd. Knowing when to drop into code is a skill, not a failure.

Covered

Nodes, items and how data flows between stepsBranching, filtering and looping without spaghettiTransforming and merging data shapesExpressions, variables and the code noden8n vs Make: hosting, pricing and when each winsVersioning a workflow and testing before it runs live

Connecting anything: apps, APIs and the ones with no integration

The four hundred ready-made connectors cover the easy half. The interesting work starts with the tool your company actually uses, which has an API and no node. You will authenticate against it by hand, read its documentation, and wire it in — after which no tool is ever off-limits again.

Covered

HTTP requests, headers and payloads without fearAuthentication: keys, OAuth and where credentials liveReading an API reference you have never seenWebhooks in both directionsPagination and large result setsRate limits and backoff

Where the model belongs — and where it does not

An LLM step is the most expensive, slowest and least predictable box in your flow. It earns its place only on the fuzzy middle: classifying, extracting from unstructured text, drafting. You will build agentic steps that call your own tools, and you will also delete a few model calls a regex handled better.

Covered

Classification, extraction and drafting: the tasks worth a modelForcing structured output a later step can rely onAgentic steps that call your own toolsCost per run, and why a nightly loop multiplies itWhat happens when the model returns nonsense mid-flowDeciding when a regex, a lookup or a rule wins

Running it in production without babysitting it

A workflow only you can fix is a liability, not an asset. This session is about the unglamorous half: what happens when an API is down, when the same event fires twice, when a colleague inherits your flow. Errors are not edge cases here — they are the normal operating condition.

Covered

Error branches and retries that do not make things worseIdempotency: surviving the same event twiceAlerting a human, and choosing what deserves an alertLogging enough to debug a run from last TuesdayDocumenting a flow so someone else can own itReviewing cost and hours saved after it has run a month
What you leave with

Build and deploy a production-grade AI automation: a workflow running on a real trigger against your own work, with error handling, idempotency, a measured cost per run, documentation a colleague can follow, and a before-and-after count of the hours it gives back.

Offered in

Questions

Do I need to code?

No, and that is genuinely true here rather than marketing. But "no code" does not mean "no thinking": you will work with conditions, data shapes, loops and error handling — that is programming, with boxes instead of syntax. People who are comfortable reasoning about logic do well regardless of background. We do show the code node, because there is always one step where twelve visual boxes collapse into three lines, and knowing that escape hatch exists is part of the job.

n8n or Make — which should I learn?

Both, because the choice is situational and the concepts are the same. Make is faster to start with and its pricing is predictable; n8n can be self-hosted, which matters when data cannot leave your infrastructure, and it goes further when a flow gets complex. Once you understand triggers, data shapes and error handling, moving between them takes an afternoon. The tool is not the skill.

Will an automation actually save the hours people claim?

Sometimes, and honesty about when is why this module insists on measuring. A workflow you spend three weeks building to save twenty minutes a month is a loss you can dress up as progress. The wins are real where the task is frequent, rule-bound and boring — invoice routing, lead qualification, report assembly. They are illusory where the task is rare or full of judgement calls, because you end up maintaining a fragile flow that a human handled fine. Your capstone reports a before-and-after count for exactly this reason: an automation that cannot show its hours does not deserve to keep running.

Other modules