Skip to main content

Loop engineers & autonomous runs

Autonomous agents are measured as runs: cost, iterations & whether they merged.

Yardstick treats autonomous coding agents as a first-class kind. Each run is one plan-to-PR attempt, with its cost, iterations, wall-clock time & outcome (merged, abandoned, or reverted). The number that matters is cost per merged run, plus the autonomy signal: how often a run merges with no human edits. Runaway-loop guardrails flag a run that spins past a spend cap or keeps iterating without merging, so an autonomous agent never quietly runs up the bill.

Did this answer your question?