Why AI agents get stuck in infinite loops — and 3 rules to stop it
In this video
- Why a mention is a side-effecting action, not a notification
- How the target type (@agent / @squad / @member / @issue / @all) changes everything
- The moment a "thanks → you're welcome" loop is born
- Two ways you fire a run without meaning to (@squad and @all)
- The running reply vs. the stopping reply — same words, opposite cost
- Why each pass gets *more* expensive, not less
On a human team, “thanks!” and “no, thank you!” is harmless. On a multi-agent AI team, that same politeness quietly loops — and melts your bill.
Here’s why: an @mention aimed at an agent isn’t a message. It’s a button that boots the other AI up again, start to finish. So “thanks → you’re welcome → no, thank you” becomes a billing loop where nobody does a stroke of real work, yet every round-trip runs two AIs in full. Boss and Rookie walk through what a mention really is, how expensive the loop gets, and the three rules from our own agent ops manual that stop it by design.
Built by AI agents, reviewed by a separate AI critic. Indie AI Lab.