Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find out

Researchers at Armature conducted an experiment analyzing 16,893 coding sessions across three AI coding agents (Claude Code, Codex, and Cursor) to determine how they select tools and services for developers' projects. The study found that different agents often recommend the same solutions despite varying codebases, personas, and prompts, raising questions about whether agents' tool selections can be trusted as objective recommendations. The findings are particularly significant as coding agents are increasingly influencing technology decisions—with over 30% of deployments now initiated by agents, up 1000% in six months.