Notes on reviewing code an agent mostly wrote. What breaks in the workflow, what other teams are publishing about their own review load, and which tools hold up when you check the claim.
2026
-
Open source AI code review: which part is actually open
Five arrangements all get called open source AI code review. The license badge does not tell you which one you have. What to check in the repo, the license file, and the vendor's own README.
-
What an agent's fix commit does to the approval you already gave
Approval invalidation is a repository setting, and an agent's fix loop hits it on every push. What GitHub, GitLab, CodeRabbit and Kodus document about re-review rounds.
-
The review throughput ceiling that pricing pages leave out
Per-review file caps and per-developer refill rates decide whether an AI reviewer keeps up with agent output. What CodeRabbit, Copilot and Kodus document, and which limit binds first.
-
Free AI code review: what the free tier actually covers
Free means four different arrangements in AI code review. What CodeRabbit, Greptile, Qodo, Copilot and Kodus include on a free tier, and the metering unit that decides whether you get a review or a demo.
-
What your review queue's ranker can and cannot see
Once pull requests outnumber review hours, a ranking layer decides what gets read. CodeRabbit Triage, Kodus Cockpit and GitHub's own metrics each document what their numbers leave out.
-
Which AI reviewer approvals satisfy your merge requirement
A bot review only blocks or clears a merge if the repository says so. What Copilot, CodeRabbit, Kodus and Graphite document, and the effort dial behind it.
-
AI code review on GitLab: what the tier decides
Required approvals are a paid GitLab feature. What CodeRabbit, Qodo, GitLab Duo, Copilot and Kodus each document for GitLab, tier by tier.
-
How much of code review an AI reviewer actually carries
Firm-level data shows AI review tools write about a quarter of review comments. Coverage, not finding count, is the number to ask a vendor for, and here is how to read it from config.
-
Which of your rule files the reviewer actually reads
Every AI reviewer claims to read AGENTS.md and CLAUDE.md. What decides whether your standards land is which rule wins when two apply to the same file, and whether the tool can show you which one it used.
-
AI code review on GitHub: what the platform decides for you
Installing an AI reviewer on GitHub is a platform decision before it is a model decision. The trigger event, the checkout, and the token permissions set what it can read and write.
-
Where AI review saves time: the write-back, not the analysis
Review-time savings come from what an AI reviewer can do to a pull request, not from how well it reads the diff. The per-provider capability gap, and how to audit it.
-
The artifact worth reviewing when agents write the code
When an agent produces a change, the thing a team can actually review is the workflow that produced it. What that looks like in practice, with primary sources.
-
Review policy can't rely on flagging AI code
A developer tried to tell which of 102 F-Droid apps were LLM-generated and re-labeled 10% on a second pass. Detection is too weak to build review policy on.
-
What self-hosting AI code review actually costs
Augment Code's 450K-file monorepo test puts a number on self-hosting review tools: $4,100 to $9,100 a month once infrastructure and maintenance enter. Method and sources.
-
When review time plateaued, reviewers had stopped reading
Salesforce's code volume rose 30% but review time on the largest PRs flattened. That plateau is a disengagement signal, not a throughput win.
-
GitHub's official AI review guide is a mostly-human checklist
GitHub's own tutorial for reviewing AI-generated code is an 8-step checklist, and almost every step is assigned to a human reviewer. Read the primary source.
-
Multi-repo AI review: a context problem, not a volume problem
Why AI code review scales or fails in large multi-repo teams depends on cross-service context, not on how many PRs it can scan.
-
Where AI review pays: the codebase matters more than volume
The review bottleneck is not that there is more code. The gains from AI shrink on mature codebases, so where you spend review effort shifts.
-
Review time is up 441% under vibe coding. That is the real problem.
Field note on a peer-reviewed review reporting code-review time up +441% under AI-assisted development, and why the productivity number to watch is the one nobody markets.