Judge a nightly agent that upgrades dependencies

A teammate set up an AI agent that runs every night across your 120 repositories. In each one it upgrades every outdated dependency, runs the full CI, and loops on failures, up to 15 times, until the build is green. Then it opens a PR.

The model bill is now $9,000 a month, CI queues stay full until 10:00 each morning, and teams merge about 1 in 5 of its PRs. Leadership likes the idea of dependencies that update themselves.

Is the agent worth what it costs? Decide what you measure and what you change.