The Safe Way to Ship Production Code Written by AI Agents
How to safely ship AI-generated production code with permissions, testing, security gates, and review.
A new type of pull request is emerging in software development: one where a bot writes the implementation after creating a feature branch, a human approves, and the change is shipped—without the human typing any code. GitHub reports developers used the Copilot coding agent to create over a million pull requests between May and September 2025, resulting in 518.7 million merged pull requests in 2025, up 29% year over year.
The industry conflates three levels of tooling: autocomplete, assistant-style tools like GitHub Copilot Chat, and agentic tools that plan multi-step tasks and execute them without human approval of each step. The shift from autocomplete to agentic tools has changed the risk profile of AI in coding, making it necessary to review the output of an autonomous process, not just a suggestion.
However, productivity gains from AI tools are contested. METR, an independent AI evaluation nonprofit, found that experienced developers were 19% slower when using AI tools, contrary to expectations of a 24% speed increase. This finding was due to a small, experienced cohort and potential bias from developers who felt less productive without AI.
Furthermore, many developers now resist working without AI access, complicating measurements of true productivity gains. Instead of focusing on individual task completion speed, organizations should measure shipped, working software across teams, as AI-assisted teams showed a 21% higher task completion rate and 98% more merged pull requests.
Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
