Anthropic tries to hide Claude’s AI actions. Devs hate it • The Register

February 16, 2026 by Cnznews

Anthropic has updated its AI coding tool Cloud Code, changing the progress output to hide the names of files that the tool reads, writes, or edits. However, the developers pushed back saying that they needed to see which files were accessed.

Version 2.1.20 collapsed the output so that instead of showing, for example, the file name and how many lines were read, it would only print “Read 3 files (Ctrl + O to expand)”, according to a post complaining that “cloud code is being minimized.” Full details can still be obtained with a keyboard shortcut, but using it constantly is annoying and impractical.

Developers have many reasons for wanting to see file names, such as for security, to instantly know if the cloud is pulling references from the wrong files, and to have an easy audit of past activity by scrolling through conversations. One person wrote, “When I’m working on a complex codebase, knowing what context the cloud is pulling helps me catch mistakes quickly and move the conversation forward.”

There is also a financial impact. If developers sense that the cloud is going down the wrong path, they can intervene and avoid wasting tokens.

A GitHub issue on this topic came with a response from Boris Cherny, creator and head of cloud code at Anthropic, that “This is not a Vibe coding feature, it’s a way to simplify the UI so you can focus on what matters, interoperability and bash/mcp output.” He suggested that developers “try it for a few days” and said that Anthropic’s own developers “appreciated the reduced noise.”

Cherney said developers who want more details can enable verbose mode. Responses were tepid, with one person writing: “Verbose mode is not a viable option, it has too much noise.”

Another observation was that the new default output, such as “Discovered 2 patterns, read 3 files,” does not provide any useful information. One user said, “This is not good simplification, this is foolish removal of valuable information.”

Cherney responded to the feedback by making changes. “We’ve reworked the existing verbose mode setting for this,” he said, so that it “shows file paths for read/search. Does not show completion, hook output, or subagent output (coming in tomorrow’s release).”

The problem with this is that making the verbose mode less verbose is a bad change for people who wanted full details.

Cherney also participated in a lengthy discussion on Hacker News. “The cloud has become more intelligent, it lasts longer, and it’s able to use more tools more agentically… the amount of output it generates can become increasingly overwhelming in the terminal, and that’s something we often hear from users,” he said.

Users who want concise output seem to be mostly absent from the discussion. One response said, “I can’t tell you how many times I benefited from looking at the files the cloud was reading to understand how I could step in and give it a little more context… saving thousands of tokens.”

Cherney said the revamped verbose mode was the solution, and cloud code would still default to the concise view.

The debate is important because if AI tools like Cloud Code hide from developers (or other users) what they are doing, mistakes are more likely to be made. Another developer said, “I’m a cloud user who has recently become frustrated with how opaque the system has become.” “Right now Claude can’t be trusted to get things right without constant inspection and frequent corrections, often just for one step. For people like me, it’s make or break. If I can’t follow the logic, read the intent, or catch the logic quickly enough, the session simply ends through my token quota.”

Cloud code changes frequently, so it’s likely that this aspect will be changed further, but there is no indication yet that it will revert to the old behavior. ®

<a href