ChatGPT Work vs Claude Cowork: One made me nervous testing file organization
ChatGPT Work found duplicates and saved time, but missing permission prompts raise safety concerns.
David Gewirtz of ZDNET compared OpenAI's ChatGPT Work with Anthropic's Claude Cowork for desktop automation, specifically organizing PDFs in a Downloads folder copy. ChatGPT Work, released last week, analyzed 447 PDFs, identified duplicates by comparing content (not just file names), and removed them. It used 11% of the $20 Plus plan's capacity. However, Gewirtz noted that Work lacked permission prompts before acting, making it feel less safe than Claude Cowork, which had stricter controls. The test revealed that while Work saved time on tedious cleanup, its agentic behavior without explicit user permission for each action is a major deal breaker.
Despite finding duplicates that Claude Cowork missed and providing detailed folder summaries, ChatGPT Work did not flag generically named files like Cowork did. Gewirtz highlighted that missing permission prompts are the biggest concern, as agentic AI could go rogue without proper safeguards. For now, he is not willing to let Work loose on other desktop projects, emphasizing that safety features like per-action approval are critical for professional use.
- ChatGPT Work used 11% of a $20 Plus plan to organize 447 PDFs, finding duplicates by content comparison.
- Claude Cowork felt safer due to explicit permission prompts, which Work lacks.
- Work identified duplicates that Cowork missed but failed to flag generically named files.
Why It Matters
For professionals, agentic AI must balance efficiency with safety; missing permissions could lead to unintended data actions.