Skip to content
You Need To Understand This

Stop doing the same thing twice

Automation

Most people automate the thing that annoys them rather than the thing that costs them, build it in whatever tool someone recommended, and find out eighteen months later what that choice was actually costing. This module goes in the other order: work out whether the task deserves automating at all, map it properly before opening any tool, understand what it takes for two systems to talk to each other, choose a category of tool with the trade-offs in front of you, and know the narrow set of jobs where an AI agent beats a plain workflow.

5 lessons1 hr 36 min

Easier if you have done: Building Things

Lessons

  1. 01

    Should this be automated

    The arithmetic that tells you whether automating a task will save time or quietly cost you more than doing it by hand.

    You keep: Can decide before building whether a task is worth automating, and recognise the tasks that should never be.

    18 min
  2. 02

    Trigger, condition, action

    The three-part model behind every automation ever built, and how to map a manual process onto it before you choose a tool.

    You keep: Can break any manual process into trigger, condition and action, and spot the parts that will not fit.

    18 min
  3. 03

    What an API actually is

    One program asking another for something, in a format it can use — requests, responses, JSON, keys, limits and webhooks, without jargon.

    You keep: Can read API documentation well enough to say whether two systems can be connected, and what it would need.

    20 min
  4. 04

    Automating without writing code

    The real categories of no-code and low-code automation, what each is genuinely good at, and what each one costs you.

    You keep: Can choose the right category of automation tool for a task, and name what they give up by choosing it.

    20 min
  5. 05

    When an agent is the wrong answer

    What an AI agent actually is, the narrow set of jobs it does better than a fixed workflow, and the failure modes that make it dangerous everywhere else.

    You keep: Can tell whether a problem needs an AI agent or a deterministic workflow, and design the approval gates if it needs an agent.

    20 min