The AI Record
cortex hit exactly this. The moment we ran Claude Code at full throttle, writing speed jumped by an order of magnitude or more. Meanwhile the human time available to read and approve PRs only grew linearly. If the reviewer (=me) took a day off, the whole org stalled -- a classic single point of failure.

Ryosuke TsujiCTO · Executive, airCloset

Post: Human-on-the-Loop: AI Reviewing AI PRs at cortex -- 769 PRs/month while raising the quality bar, DEV Community, May 26, 2026, at How the review bottleneck stops forming.

Checked against the source [ Grade B ] Said in the last year

Filed under

Who is speaking

Ryosuke Tsuji, CTO. Writes as CTO of airCloset and says he designed and operates the cortex systems the post describes.

How this was checked

Checked against the source by a first reader, then confirmed independently by a second reader who fetched the source without seeing the first reader's notes (Independent review agent (Claude Sonnet), blind second pass from passage text and citation only, no repository access, 2026-11-06).

Found in the bottleneck section, including 'If the reviewer (=me) took a day off'. Exact match apart from spacing around markup. The author's own words.

How to cite this passage

Ryosuke Tsuji (airCloset), “Human-on-the-Loop: AI Reviewing AI PRs at cortex -- 769 PRs/month while raising the quality bar”, DEV Community, 2026-05-26, at How the review bottleneck stops forming. https://dev.to/ryantsuji/human-on-the-loop-ai-reviewing-ai-prs-at-cortex-769-prsmonth-while-raising-the-quality-bar-4lh5 Archived at https://the-ai-record.com/passages/aircloset-moment-we-ran-claude-code-full-2026

More from the same source

cortex went all-in: the AI writes the code, the AI reviews the code. What humans keep their hands on is "tuning the prompts and guidelines themselves" -- not making decisions inside each individual PR, but watching the system from above and adjusting.

Ryosuke TsujiCTO · Executive, airCloset

Post: Human-on-the-Loop: AI Reviewing AI PRs at cortex -- 769 PRs/month while raising the quality bar, DEV Community, May 26, 2026, at cortex's answer: move the reviewer role to AI as well.

Checked against the source [ Grade B ] Said in the last year Passage page

Every one of those 769 PRs had an AI reviewer as the first reviewer, with an average of 10.8 review-fix loop iterations per PR (max 56). 1 in 5 merged within 10 minutes, roughly half within 30 minutes. What humans do now is look at review outcomes and tune the review prompt and the guidelines themselves -- this is human-on-the-loop, not human-in-the-loop. Humans operate on the policy layer, not the execution layer.

Ryosuke TsujiCTO · Executive, airCloset

Post: Human-on-the-Loop: AI Reviewing AI PRs at cortex -- 769 PRs/month while raising the quality bar, DEV Community, May 26, 2026, at Start with last month's numbers.

  • 769 pull requests prs merged, 30 days, Apr 21 to May 21. No person in the loop
  • 10.8 iterations per PR avg review iterations, 30 days, Apr 21 to May 21. No person in the loop

Checked against the source [ Grade B ] Said in the last year Passage page

When the reviewer's machine receives an event, the script spawns claude -p and walks through 9 dimensions (Graph / Architecture / Security / Test / Doc / Impact / Observability / AI-Antipattern / Recurrence) sequentially, then reads the verdict marker the AI emitted at the end and posts APPROVE or REQUEST_CHANGES via gh pr review.

Ryosuke TsujiCTO · Executive, airCloset

Post: Human-on-the-Loop: AI Reviewing AI PRs at cortex -- 769 PRs/month while raising the quality bar, DEV Community, May 26, 2026, at How the auto-review system is wired.

Checked against the source [ Grade B ] Said in the last year Passage page

Something wrong? Report a correction. JSON: /api/v1/passages/aircloset-moment-we-ran-claude-code-full-2026.json