Summary
Sachin Gupta, a software engineer at eBay, presents ReviewDebt — a quantitative framework for measuring the growing gap between AI-generated code volume and the human review attention it receives. The premise is grounded in hard numbers: GitHub’s October 2025 report shows commits up 25% year-over-year while PR review comments dropped 27%. A Faros AI 2026 benchmark tracking 22,000 developers across 4,000 teams found median PR review time has increased 441.5%, reviewed PRs now take 5.4 times longer than before, and 31% more PRs are merged with no review at all.
Gupta introduces five signal families that compose a ReviewDebt score from 0 to 100: AI authorship detection (co-authored footers, Copilot/Codex branch-name prefixes), evidence and rationale gaps (does the PR explain why, not just what), test theater (assertions that verify what the code does rather than what it should do), architectural drift, and reviewer fatigue indicators. He scores three real pull requests side-by-side and presents a cross-repo scan of 524 PRs from three anonymous public repositories.
The score maps to four action bands: 0–24 (low burden), 25–49 (standard review), 50–74 (author must provide evidence before senior review), and 75+ (high risk, block until addressed). Gupta recommends teams back-calibrate the default signal weights against their last 200 merged PRs so that the score aligns with their own engineering instincts before adopting it in CI pipelines.
📺 Source: AI Engineer · Published July 12, 2026
🏷️ Format: Deep Dive
