Automation is a funny thing.
Everyone knows it saves time, but it is notoriously hard to get right. And the difficulty does not rise in a clean linear way. It compounds. Every human judgment step makes the automation harder in ways that are out of proportion to the task.
Take LinkedIn commenting and outreach.
Technically, it is not difficult. Log in. Find a post. Leave a comment. Send a DM. This is automated constantly — by software, AI, and human VAs.
But automating it well is extremely difficult.
You have to choose the right person. Then the right post. Then the right kind of comment. Sometimes the right comment is funny. Sometimes it is sharp. Sometimes it is generous. Sometimes it should not be left at all.
You also need to know who you are talking to. I have had people reach out on behalf of someone I had already worked with. Technically, the automation fired. Socially, it failed.
The hard part is not the click path.
The hard part is the judgment.
A good automation needs to extract the details that live only inside a person’s head: the small decisions they make twenty at a time without even noticing. The literature calls this tacit knowledge.
That tacit knowledge is everywhere. It is how a founder qualifies a lead. How a recruiter knows a candidate is wrong even though the resume is right. How a designer knows a page feels cheap. How a salesperson knows when to push and when to back off. How an operator knows which edge case matters and which one is noise.
If you skip this layer, AI just makes the bad process faster.
If you extract it, the automation has a chance to become useful.
So before you automate a workflow, slow down and ask: where are the human judgment calls hiding?
Write them down.
Name the patterns.
Find the exceptions.
Only then should you build the machine.