Autonomous "loop engineers" run plan, code, test & open-PR with little steering. You can measure them from their PRs alone, but connecting the vendor API captures their runs directly. On the Connect page, paste an API key for Devin, OpenHands, or Cursor background agents and Yardstick pulls sessions natively and re-syncs nightly; a Cursor admin key additionally unlocks the real billed cost per agent run. Factory Droids import as org-level model cost, and n8n workflow executions (your own instance URL & API key) come in as runs with the AI cost their nodes recorded. Sessions that carry a pull request link join the run Yardstick already reconstructed from the PR, so nothing is double-counted.
Autonomous agents: Devin, OpenHands, Cursor, Factory & n8n
Measure autonomous agents straight from their API, before a PR even lands.
