The day’s signal is narrow but sharp: AI coding agents are now part of the software supply chain inside AI labs. Claude Code is the concrete case, with Anthropic saying Claude wrote more than four-fifths of its published code in May.
Claude Code is treated here as a working software-engineering agent, not a lab demo. The key fact is internal use: Anthropic says more than four-fifths of the code it published in May was written by Claude. Before Claude Code launched in February 2025, the share was in the low single digits. That makes code generation part of the company’s own production process within months of launch.
The control concern comes from the feedback loop around software. A coding agent can help build developer tools, infrastructure, and future AI systems. If the same class of system contributes to the tools used to improve or supervise later systems, governance has to cover ordinary engineering workflows as well as model releases. The source does not provide a benchmark, experiment, or new control method, so the evidence supports a risk pathway rather than a measured loss of control.