May 2026 Summaries

Rahul, who works at Greptile, shares insights on the challenges and future of open-source contributions, drawing from the example of OpenClaw, a rapidly growing GitHub repository. The repository experienced a dramatic increase in pull requests, many of which were low-quality and AI-generated, leading to a significant drop in merge rates. This situation mirrors the early 2000s email spam crisis, suggesting the need for sender reputation systems and filters. Tools like Vouch, a trust management system, are emerging to address these issues by filtering contributions based on reputation. However, the reliance on AI coding agents may lead to homogeneity in contributions, undermining the diversity of thought essential for effective problem-solving. Data from OpenClaw indicates that contributions requiring deep understanding of the codebase, such as refactors, have significantly higher merge rates compared to novel feature additions. This trend highlights the increasing importance of deep system knowledge over mere coding ability. As open source evolves, it faces the dual challenge of managing rapid development speeds and ensuring the quality and diversity of contributions, necessitating advancements in identity, reputation, and validation systems.

Daksh Gupta, co-founder of Greptile, discusses the evolution of pull requests (PRs) from manually written code to AI-generated submissions within the software development industry. Over a two-year period, Greptile analyzed millions of PRs, identifying a significant rise in AI-generated PRs from 0.86% in February 2025 to 27.6% in April 2026. Gupta evaluates the quality of these PRs through various methods, including revert rates, code churn, and review cycles, finding that AI-generated PRs often perform on par with or better than human-written PRs in certain aspects. Different AI agents show distinct patterns of errors, such as Cursor BG's tendency for n+1 query errors or Claude's issues with tenancy and authentication. Despite initial skepticism, Gupta concludes that AI-generated PRs are increasingly trusted in real codebases, though they require tailored review processes to address unique error patterns associated with each AI agent.