Too Much Work Reached the Founder, So the Agents Now Count It Every Week
1What happened
Our founder runs a team of AI agents. In one week (Oct 14 to 21) he kept getting pulled into work the agents should have finished alone. Sign-ins got stuck and needed his help to finish. Updates came as long walls of text he had to read twice. Posts to him were full of internal numbers and code names, about 14 a day. Each one cost him a few minutes. Together they made the agents feel like more work, not less. First we added a new rule after each complaint: "keep it short", "no jargon". Each held for a day or two, then slipped, because nobody counted. Then we let the agents fix what they noticed. They fixed what was easy to see, not what cost him the most time.
2What we did
Once a week, on a set day, the agents read the founder's own messages from the last 7 days.
They count every time something reached him that shouldn't have.
They pick the top 3 by count. Each gets one fix, an owner, a due date, and a check that proves it worked.
This week's 3: sign-in steps that read back every field before they submit; a check that refuses a message to him with a paragraph over about 300 characters; a warning when a message to him has internal numbers or code names.
The count is the scoreboard. A fix that doesn't lower the count next week gets replaced.
3How we know it worked
Not yet proven. The first scan ran Oct 21 and found the 3 problems above. Next week the same count runs again and is compared line by line. We'll add the result here.
4What you can take from it
If your AI agents feel like more work, count what reaches you for one week. Fix the top 3, and count again. One-off rules fade; a weekly count doesn't.
In our glossary: Count and check
