Claude Code's binary reveals silent A/B tests on core features
Summary
The article alleges silent A/B testing on Claude Code that degrades user workflow. It identifies a GrowthBook test tengu_pewter_ledger with variants null, trim, cut, and cap, noting that the cap variant severely restricts output and is logged via telemetry without opt-in. It advocates for transparency, configurability, and human-in-the-loop governance for AI tools used in professional settings.