AI · Jan 30, 2025 · 3 min read
Notes on Automation
Automation is not about replacing people. It is about changing what people spend their time on.
Automation is not about replacing people. It is about changing what people spend their time on.
This distinction matters. When businesses approach automation as a headcount reduction exercise, they usually get it wrong. They automate the visible parts of a job and discover that the invisible parts - the judgment, the relationships, the handling of exceptions - are still being done by humans, just with less support.
When businesses approach automation as a time-reallocation exercise, they usually get it right. They identify the work that is repetitive, rule-based, and cognitively draining, and they automate it. This frees up the people doing that work to focus on the things that actually require human judgment.
The result is not fewer people. It is more capable people.
What is worth automating
Not everything is worth automating. The economics need to make sense, and the reliability needs to be high enough that the automated version is actually better than the human version.
I think about automation candidates in terms of three questions:
Is it repetitive? If someone is doing the same thing more than a few times a day, it is worth asking whether it needs to be done by a human at all.
Is it rule-based? If the decision can be described as a set of rules - even complex rules - it can probably be automated. If it requires genuine judgment, it probably cannot.
Is it critical? The higher the stakes of getting it wrong, the more reliability you need before you automate it. Some things are worth automating even if the error rate is 5%. Others need to be right 99.9% of the time before you trust them to a machine.
The reliability problem
The hardest part of automation is not building the system. It is making it reliable enough to trust.
This is where a lot of AI-powered automation falls down. The model works most of the time. But "most of the time" is not good enough for operational systems. You need to know what happens in the cases where it does not work. You need fallbacks. You need monitoring. You need a way to detect when the system is failing and route those cases to a human.
Building reliable automation is slower and less exciting than building impressive automation. But it is the only kind that actually changes how a business operates.
Where I am
I am building in this space. Learning what works, what does not, and where the real leverage is. These notes are part of that process - thinking out loud as I figure it out.