theregister.com
|
ksl
|
|
Anthropic shipped a Claude Code update that collapsed file access details behind a summary line – “Read 3 files” instead of showing actual paths. Developers reacted badly and fast. The core complaint is practical: when Claude pulls context from the wrong file, you need to see it happening in real time, not discover it after burning through tokens. Auditing past sessions also breaks when the activity trail is condensed. Anthropic’s Boris Cherny framed the change as noise reduction, but the gap between “simplified UI” and “obscured behavior” is exactly where trust erodes. The partial fix – repurposing verbose mode to show paths – satisfied almost nobody. Transparency concerns around AI coding agents keep surfacing across tools from Cursor to Copilot, and each time the pattern is the same: convenience defaults that strip visibility from the people who need it most.
