Research from the review queue.
Benchmarks, field reports, and short cheat sheets. We measure what actually changes when AI code review goes live.
Featured.
State of AI vs. Human Code Review
We analyzed 24,000 PRs across 18 customers. AI-written diffs produce ~1.7x more issues per LOC, but humans still miss the same 12 bug classes 38% of the time.
Pre-merge gates: a quantitative case for blocking
Comparing teams that block on AI review vs. teams that advise. Spoiler: the blockers ship faster, not slower — once they get the rule profile right.
10 prompting tips for better AI-assisted code
Field-tested patterns from teams using PRSentinel + Copilot/Cursor side-by-side. Same model, different ergonomics.
All articles.
Reverse tunnel architectures for code review on private networks
How PRSentinel runs in air-gapped environments without inbound firewall rules.
Policy-as-Code for AI coding agents
The missing layer: how rule profiles encode org standards into LLM-aware policy.
Agentic CLI commands every team should know
Pre-commit hooks, PR readiness checks, and migration assistants — all from the terminal.
Does AI code review hallucinate? A measurement framework
Replicable benchmarks for false-positive and false-negative rates across 6 LLM providers.
Migrating a monorepo to PRSentinel in a weekend
Field notes from a 1.2M-LOC migration: profile authoring, false-positive triage, rollout.
Want one of these in your inbox?
A short newsletter — once a month, no spam.