Automating work that should not exist

The most expensive thing AI does is let you automate work that should not exist.

Somewhat famously Elon Musk has a five-step process for improving anything. First, question the requirement, then delete the step if you can. Next, simplify what remains and speed up the cycle. Only then, automate. His real point is about order, not the steps themselves. The most common mistake engineers (and now increasingly non-engineers!) make is improving something that should not exist at all.

AI makes the last two steps almost free. Speeding up and automating now cost almost nothing. That is exactly what is worrying. Automation used to be expensive, so people thought hard before they built it. Automation is cheap now. People build first and ask questions later, if they ask at all.

But it has some real cost: Automation of a bad workflow is still bad, and it likely precludes you from thinking about an alternative flow that is better.

The cheaper automation gets, the more real value sits in deletion. Before you let AI speed something up, ask whether it should exist at all. That question still deserves your time. The automation itself barely needs any now.