A hard limit on how many iterations a given automated routine or agent loop is allowed to run, meant purely to prevent runaway/unbounded execution.
Apply: add an explicit maximum-iteration setting to every automated routine or agent loop before letting it run unattended. The source offers a combined prompt that sets this cap alongside the Minimum Viable Model (MVM) choice for a skill in a single pass.
Motivating anecdote: a client already at step 3 of the maturity ladder, with guardrails supposedly already in place, woke up to find AI had burned through over $3,000 in tokens overnight on a single task because the loop had no iteration cap.
"one morning he woke up and AI burned through over $3,000 worth of tokens on a single task"
The point drawn from this: reaching a later maturity stage does not by itself prevent runaway cost — the specific missing control (here, the absence of a max-iteration cap) is what mattered, not how far along the Boris Cherny's 4-Step AI Maturity Framework the operator otherwise was.