Harvard researchers Fiona Chen and James Stratton analyzed granular engineering analytics from Jellyfish - about 300 million work events across more than 700,000 engineers at 700+ firms from 2021 to March 2026 - to measure the operational impact of AI coding assistants and autonomous coding agents. Using direct measures of AI adoption and GitHub activity with difference-in-differences regressions, they found that introducing agentic coding tools sharply raises developer output metrics: roughly 30% more lines of code, 20% more commits, and 23% more pull requests on average.
Those upward shifts did not translate into greater software delivery or reduced headcount. Jira-tracked issue and epic resolution rates showed no statistically significant change. Instead, review work ballooned: time from pull request submission to merge rose about 49%, the share of PRs requiring changes nearly doubled, comments per PR increased ~35%, and 14% more workers were doing reviews. AI-driven review remains limited - by March 2026 agents authored about 23.3% of review comments and handled 10.8% of PRs - so human reviewers remain the bottleneck. The net effect is that coding efficiency gains are largely absorbed by downstream review costs, leaving firms with more code but not more completed software.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
![Martin Odersky – Keynote: The End of Programming? [video]](https://hntoday.pixelshot.io/screenshots/20261012/e928e8e31c694fd6_tn.jpeg)
Martin Odersky discusses the future of programming and whether traditional coding will become obsolete. He explores advancements in programming languages and tools that could transform software development. (youtube.com)
Codex CLI's persistent, self-updating background daemon is now enabled by default. The feature is intended to improve performance but is now available as an opt-in setting to give users more control. (github.com)
Sindre Sorhus rewrote his website using Swift, adopting SwiftUI-like components, styles as modifiers, and a compiler that detects broken links before deployment. This approach streamlines development by using a familiar language and leveraging compile-time checks for web page consistency. (sindresorhus.com)
A developer created a modern web application in C using CGI, SQLite, and minimal dependencies, proving CGI's relevance in 2026. The method focuses on core HTTP concepts and simplicity, offering easy deployment and security advantages. (prophitt.me)
Daniel Lemire demonstrates a method to parse JSON data at tens of gigabytes per second using Python, emphasizing performance optimization. This approach significantly improves JSON processing speeds in data-intensive applications. (lemire.me)
Lukáš Lalinský developed zt, a type-safe HTML templating library for Zig, addressing the need for safe, fast, and HTML-like syntax in low-level web development. Inspired by JSX and Templ for Go, zt compiles templates into Zig code, enabling safe and efficient web output without runtime errors or allocations. (lalinsky.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.