A Reddit X20 user runs nine scheduled ChatGPT agents across a repo—PR bugs, perf sweeps, docs clarity, unfinished features—each posting to its own GitHub issue as a message board, with human approval before merges.
On HN, joshgachnang says Cursor Cloud Agents keep multiple workstreams going for his team, and computer use lets agents actually test features and show they work on PRs—alongside Bugbot as their best AI reviewer so far.
An HN commenter described a Claude Code setup with reusable skills for a UI library/Storybooks/APIs, a post-change review agent, PR checklist reviews, scheduled docs/coverage agents, and ticket triage that leaves a halfway-done plan on each issue.
Amplitude runs 1,000+ unprompted Cursor agent runs a week: Slack bugs become tickets and fixes, Bugbot merges 60–70% of low-risk PRs, and weekly production commits tripled with Cursor among top-3 committers.
AI Scan is GA: it complements CodeQL by auto-scanning PR diffs for languages/frameworks CodeQL does not cover (PHP, Shell, Terraform HCL, Dockerfiles, JSP, Blazor, etc.), surfaces advisory findings with explanations and suggested autofix, and consumes Code Security AI Credits without blocking merges.
Anthropic’s PR Code Review left research preview: auto-reviews on new GitHub PRs in ~6–10 minutes, with Quick/Standard/Deep intensity per repo for Team and Enterprise.
Anthropic’s Claude Code team documents a layered review stack—pre-PR verify, AI review + security block, risk-tiered human approvals, merge queue/flags, and fast rollback—so most agent PRs merge with ≤1 human while revert rate fell >40% as volume doubled.
Bitbucket Cloud pattern for large monorepos: a matcher agent maps changed paths to the right review docs via Bitbucket MCP, then a fresh reviewer comments only against that short list—humans still approve.