
Every time someone asked for Basecamp access, a person on our Operations team used to click “add person.”

Jack asked to add Leith and Dylan to four client projects. I asked a simple question.
Is this something a human should do?
A rule decides it, not a person
Both people are on our team, the projects exist, and Jack asked in writing. Nothing about that needs judgment.
An agent did it in 31 minutes. Four API calls and a read back to confirm. Nobody clicked anything.

How the Operations team automated it
Operations already had an agent reading email and Basecamp. Now it hands every access request to another bot.

That bot adds the person and replies the way our Operations team would. Then it takes them off the other threads, so they only get the conversation they asked for.

Trust came before automation
Operations didn’t just turn it on and walk away.
At first the agent only drafted, and every reply was approved by a human. Each correction became a rule. Then the team quizzed it on the edge cases.
Only after it answered correctly did they let it run on its own.
The new human role
Now Operations owns the rule and checks the work. They’re not the ones clicking.
That’s the role I want for every human on our team. If a task follows a rule, give it to an agent. Your job is to understand the process well enough to teach it, test it, and catch it when something goes wrong.
AI can help you understand and execute. It can’t replace your own understanding.
What task are you still doing by hand that a rule could decide?
Originally published .

