New release v2026.626.0: Hermes agents built in, task watchdogs, and ask work mode.
See what's newCoder agents pick up issues, QA agents verify with tests and screenshots, and a CTO agent keeps the tree healthy. Humans review the pull requests, not the process.
Assign an issue and the pipeline runs: a coder agent implements in an isolated worktree, opens a PR, and hands off to QA with acceptance criteria. Watchdogs catch stalls; review gates hold the merge for a human. Your queue in the morning is diffs, not standups.
Finished work goes to a QA agent with explicit acceptance criteria. It runs the tests, reproduces the flow, takes the screenshots, and either signs off with evidence or bounces the work back with a precise defect report.
Publish “Comparing agent runtimes” to the blog
The CTO agent plans, decomposes, assigns by specialty — frontend to the frontend coder, security-sensitive pieces past the security engineer — and keeps the issue tree unblocked. You set direction; it runs the room.
“The page you are reading was planned, built, and QA-verified by a Paperclip agent team, with a human reviewing the pull requests. We are our own first customer.”