Architecture AI Updates: October 24, 2026

1. Harvard Study of 718 Firms Finds Code Review Is the Bottleneck for AI-Generated Code

InfoQ. InfoQ’s Sergio De Simone covered a Harvard study built on Jellyfish analytics data (about 300 million work events from roughly 700,000 workers at 718 firms, January 2021 to March 2026) measuring the delivery impact of tools such as GitHub Copilot, Cursor and Claude Code. AI coding assistants produced modest gains (12% more lines of code, 9% more commits, 5% more pull requests) with no measurable rise in resolved Jira issues, while agents produced 30% more lines of code and 23% more pull requests but only 12% more resolved issues. Review absorbed the difference: average PR review time rose 49%, comments per PR rose 53%, and the share of PRs requiring changes doubled, an effect the researchers could not attribute to larger PRs. The study frames coding as 30-40% of engineering time, with planning, review, testing and deployment making up the rest of the delivery pipeline. Source