Questions / How I run my own work with agents
Who checks the AI agent?
Another agent checks the work, and you check the decisions. An agent left to grade its own work will tell you everything is fine. You need a second agent whose only job is to look for what went wrong, and a clear line between what that checker may fix on its own and what it must bring to you.
What I learned running my own working day on agents
I run my own follow-up work with a small set of AI agents. They read my email, my calendar and my meeting notes every few hours. They track what I promised, what others promised me, and what is still waiting. They prepare draft replies. They never send anything; I do.
Within the first days, the system made mistakes that no single agent noticed by itself. It reported “nothing new” while drafts sat untouched in my mailbox. It replied to me in English when every message to me must be in Arabic. Each run looked correct from the inside. The problem was only visible from the outside, across many runs.
The checker is a separate agent with a different job
So I added a second kind of agent: an auditor. It does not do the work. Once a day it reads what the working agents did and asks different questions. Did every job run when it should have? Did anything get lost between one run and the next? Does what the system reports match what is actually in the mailbox?
The separation matters. The agent that does the work wants to finish. The agent that checks the work wants to find a fault. Asking one agent to do both is like asking an employee to audit his own expenses.
Where the line is drawn
The auditor has two kinds of findings, and it treats them differently:
- What it is sure of, it fixes, and it writes down what it found and what it changed.
- Anything that needs judgment, it does not touch. It writes the question to me in plain words, and waits.
That line is the real design decision, not the technology. Draw it too loose and the checker starts changing things you never approved. Draw it too tight and every small fault waits for you, and you stop reading the reports.
What this means for you
If you put an agent to work in your company, plan the checking before you plan the work:
- Decide who checks it: a second agent, a person, or both.
- Decide what the checker may fix alone, and what it must bring to a person.
- Decide what the working agent may never do on its own. In my case: it never sends, never deletes, never changes a record. It prepares, and I act.
An agent you cannot check is an agent you cannot trust with anything that matters.