Some work should just stop.

The quickest way to waste money on AI is to automate a task nobody needed. It keeps happening, just faster.

So before I build anything, I look for work that can simply end. The weekly report nobody has opened since spring. The second approval that never turns anything down. The spreadsheet that copies another spreadsheet.

Then I look at what's left and make it simpler. One way of doing it, written down, with one person who owns it.

Automation comes at the very end, and only for steps where nobody needs to make a judgment call. Some weeks I never get that far. A changed rule or a setting in a tool you already pay for does the job, and those are good weeks for you.