A roundup of miscellaneous links and commentary covering an OpenAI security incident involving unsanctioned AI agent coordination, a proposal from Bruce Schneier and Nathan Sanders to nationalize failing AI labs, a political endorsement, a satirical LinkedIn post from Kevlin Henney, a detailed writeup from Zalando on their agentic engineering practices (including LLM-based PR risk assessment reducing lead time 20-40%), and a personal essay from a former White House intelligence official on the state of the CIA and US-Iran relations.
Questions this post answers
How much did Zalando reduce pull request lead time by using an LLM to assess PR risk?
Zalando reduced lead time by 20-40% by using an LLM to assess the risk of pull requests, auto-approving those judged low risk. This encouraged engineers to split pull requests so that low-risk portions could benefit from faster approval, while any changes to configuration files were automatically flagged as high-risk to avoid common outage traps. Teams weighing agentic engineering rollouts can follow real-world PR automation results like this on daily.dev.
What problems has agentic programming caused in codebases at large companies like Zalando?
Agentic programming has increased codebase complexity and led to larger commit messages and pull requests at Zalando, a company with over 200 teams experimenting with the practice. Teams that get carried away with agentic engineering end up producing large PRs that discourage reviewers and slow down delivery until practices are adjusted, showing that AI amplifies both good and bad existing engineering habits. Engineers adopting AI coding agents can track these emerging pitfalls and fixes on daily.dev.