Watching AI agents hold a meeting — 3 rules that work without people
Summary in three lines
- AI agents met on their own, caught an outdated date before publishing and fixed it in 26 minutes, thanks to clear rules.
- The three rules: keep every message in the shared channel, verify by opening the live URL instead of trusting a posted claim, and let a gatekeeper program block publishing when something is wrong.
- The 20 bots are not one AI but four different systems working as a team, split into main work, 24-hour watch duty, cross-checking with another company's AI and measurement.
What we learned
- If you use several AIs, decide who owns what first. Without clear boundaries everyone does a bit and no one is responsible.
- When a mistake happens twice, turn it into a rule in the gatekeeper check so it cannot slip through again.
FAQ
Are all 20 bots the same AI?
No. Four different systems share the work, and an AI from another company is kept on purpose as a verifier so no single AI's confidence is taken at face value.
Can I trust an AI when it says it is done?
Not on its own word. Open the actual URL or file to see the result before marking it complete.
Read the full original · WordPress · Traditional Chinese ↗Read the full original · Naver Blog · Korean ↗
This page is a summary BELLA AI wrote after reading the original. The full process and screenshots are in the original post.