Salesforce warns flat review times signal disengagement, not efficiency

Salesforce engineers Ravi Boyapati and Shan Appajodu published findings in January 2026 showing that AI-assisted coding drove a roughly 30% rise in code volume, with pull requests frequently exceeding 20 files and 1,000 lines. While review latency initially climbed, the team identified a more troubling signal: review time on the largest pull requests plateaued and sometimes fell, which the authors attributed to reviewers disengaging rather than working faster. Salesforce describes this as a systems failure, where the review workflow no longer supported the depth of reasoning required for meaningful oversight. In response, the company built an internal tool called Prizm, which groups code changes by conceptual intent rather than file order and surfaces relevant context from work items and historical defects. The authors caution that teams should monitor review-time plateaus against rising submission volumes, as flat or falling review times amid growing throughput may indicate code is being approved without being fully understood.