Anthropic is set to make programming with Claude Code even more automated, as it will default to auto mode for Pro, Max, and Team accounts beginning August 14.
Initially introduced in March, the auto mode aims to streamline the programming process by increasing speed while maintaining control. According to Anthropic’s recent announcement, when operating in auto mode, Claude Code will advance tasks without soliciting human input at every step, except when an action is classified as “irreversible, destructive, or outside your environment.”
Testing has shown that auto mode is more effective at preventing harmful actions than manual oversight. In a study involving 1,053 paying users, auto mode identified 89% of harmful actions, while human reviews managed to catch only 13.6%. This discrepancy possibly arises from the tendency for users to consistently approve around 97% of permission requests in Claude Code.
Boris Cherny, the Head of Claude Code, shared on X that he and his team have exclusively used auto mode for several months and expressed enthusiasm about not returning to the previous permission prompts.
Additionally, Anthropic has been implementing new safety measures, which include prompt injection screening and adjustable hard deny rules, aimed at protecting against issues such as data theft.
