Anthropic tries to hide Claude's AI actions. Devs hate it
Source: Hacker News
Anthropic has updated Claude Code, its AI coding assistant, to collapse file‑access information in the progress output.
In version 2.1.20 the tool now prints a summary such as “Read 3 files (ctrl+o to expand)” instead of listing each file name and the number of lines read. The full details are still reachable via a keyboard shortcut, but developers find the extra step cumbersome and argue that the hidden information is essential for security, debugging, and token‑cost management.
Change in Claude Code output
- The previous verbose output showed file names, line counts, and other context.
- The new default output only reports high‑level actions (e.g., “searched for 2 patterns, read 3 files”).
- Users can expand the collapsed view with Ctrl + O, but doing so repeatedly is considered impractical.
Developer concerns
- Security & correctness – Knowing which files Claude reads helps catch accidental context leakage or misuse of the wrong files.
- Auditability – A scrollable conversation log with explicit file names makes it easier to review past activity.
- Token economy – Early detection of a wrong code‑generation path lets developers intervene and avoid unnecessary token consumption.
“When I’m working on a complex codebase, knowing what context Claude is pulling helps me catch mistakes early and steer the conversation.” – anonymous developer
Anthropic’s response
- Boris Cherny, creator and head of Claude Code, explained that the change is meant to “simplify the UI so you can focus on what matters, diffs and bash/mcp outputs.”
- He suggested trying the new view for a few days and noted that Anthropic’s own engineers appreciated the reduced noise.
- A verbose mode can be enabled for more detail, but many developers reported that it still produced too much output.
“Verbose mode is not a viable alternative, there’s way too much noise.” – developer comment
- After feedback, Anthropic repurposed the existing verbose‑mode setting to show file paths for reads/searches while still omitting full thinking traces and sub‑agent output (planned for a future release).
- Cherny emphasized that the condensed view will remain the default, arguing that longer‑running agents generate overwhelming amounts of terminal output.
Discussion on Hacker News
The topic sparked a lengthy conversation on Hacker News:
-
Hacker News thread:
-
Users highlighted the practical benefits of seeing file‑access details:
“I can’t tell you how many times I benefited from seeing the files Claude was reading, to understand how I could interrupt and give it a little more context… saving thousands of tokens.” – commenter
-
Some developers expressed distrust in the opaque system:
“I’m a Claude user who has been burned lately by how opaque the system has become… Right now Claude cannot be trusted to get things right without constant oversight and frequent correction.” – developer