Let's Not Automate Everything¶

What kind of tasks can't you automate, or shouldn't you automate?

First, when does automation work?¶

Automation happens when things are repeated and things have a pattern. If there is no pattern, there is no automation possible.

A simple test: if you know which tasks come up every week, and you could outsource them to someone after giving them enough context, the task is repeatable and can be automated.

1. The knowledge lives in your head¶

Sometimes you are not able to outsource a task, because the knowledge or context you hold in your brain is so much that it is very hard to export it into quantified text or a doc. You have developed it over time.

Cooking is an example. You can write it down, but it is very hard to write it down in a quantified manner. And automation requires specific steps.

2. You can't afford even one mistake¶

There are lots of things you can't automate because you are worried about even one tiny thing, one among a million things, going wrong, and that one thing would carry a huge cost. You just cannot afford even a single tiny mistake.

Examples: the medical field, or a very high-stakes deal on an international scale.

3. The operating model isn't clear¶

Some tasks can't be automated because the operating model is not clear. This covers authority, clarity, leadership, and compliance.

Some of the signs (not an exhaustive list):

• Nobody clearly owns the task.
• The approval rules aren't defined.
• The process changes depending on who's doing it.

That means that before automation, you are not sure what the operating model of the firm is.

4. You don't need AI for it¶

This one is obvious from a technical lens. In this era, when we say automation, we usually mean AI-led automation: even where there is no button or API to trigger an action, an LLM-based agent can take the action.

But there are many places where you do not need AI. A rule-based approach is enough.