There's a dirty secret in the "automate everything" movement. Every automation you build is a liability you maintain.
That Zapier workflow connecting 6 tools? It breaks when any one of them changes their API. That custom script pulling data from three sources? It silently fails when the schema shifts. That AI agent handling customer tickets? It needs monitoring, prompt tuning, and someone who understands why it suddenly started apologizing for things it didn't do.
I've built dozens of automations over the past two years. Maybe half of them are still running. The rest broke, got replaced, or turned out to cost more in maintenance than they saved in labor.
The math people skip
Time saved per week by the automation, minus time spent building it, minus time spent fixing it when it breaks, minus time spent explaining to your team why it did something weird.
That equation goes negative faster than you'd think.
What actually works
Automate the stuff that's truly repetitive and low-stakes. Data entry. Status updates. Notification routing. File organization. These are safe because when they break, nobody dies.
For anything touching customers, money, or critical decisions? Build the automation but keep a human in the loop. Not because the AI can't handle it. Because the cost of failure is too high to find out.
The real skill
Knowing what NOT to automate is worth more than knowing how to automate everything. The best engineers I know are selective. They automate the boring stuff and protect the important stuff.
That's not a limitation. That's wisdom.