The Framework

What to automate, what not to, and how.

When I started automating tasks and workflows, I made a huge mistake trying to automate everything.

So this is the framework I use now to know what and how to automate, and when you shouldn't automate at all. Run any task down the four questions and stop at the first yes.

  1. Question 1

    Does the task happen less than once a week, and can a human handle it in under 10 minutes?

    Yes

    Don't automate it

    The build, the maintenance and the day it breaks cost more than the task ever will. Automating it feels productive and loses you time.

  2. Question 2

    Are the steps deterministic - do the same inputs always create the same outputs?

    Yes

    Automate it

    n8n · Zapier · Python

    Deterministic work is the cheapest and most reliable thing to automate. No model in the loop means nothing to second-guess and nothing to pay per run.

  3. Question 3

    Is the task frequent and clearly defined, but works with unstructured input or needs to generate language?

    Yes

    Create a skill for it

    A Claude skill

    A packaged, reusable instruction set for one repeated job. It handles the messy input a rigid workflow chokes on, while still doing the same thing every time.

  4. Question 4

    You know the goal and can define it, but the path to get there is unclear?

    Yes

    Build an agent

    Claude + tools

    Give it the goal and the tools and let it find the route. This is the most powerful and the least predictable option, so it is the last resort, not the first reach.

The line the whole map draws: automate operations, never creativity.

The map tells you what to build. I build it.

Knowing a task should be a workflow is the easy half. The hard half is seeing the four places your team retypes the same information every week, and knowing which of those is a workflow, which is a skill, and which should never have been a task at all.

That is the part I do inside companies. The last time I pointed this map at an agency, it took over 200 hours a week of coordination work off a 50-editor team.

Book a Free Discovery Call

15 minutes · No pitch · Free