Anthropic is preparing to make Claude Code more autonomous. Starting August 14, auto mode will become the default experience for users on Pro, Max, and Team plans, reducing the number of permission prompts developers need to approve while working with the coding agent.
Anthropic first introduced auto mode as an experimental feature in March. The idea was to let Claude Code continue working without asking for approval at every step, while still stopping when an action appears irreversible, destructive, or likely to affect something outside the user's environment.
According to Anthropic’s announcement, when Claude Code operates in auto mode, it proceeds with routine actions automatically and only interrupts the user when a higher-risk action requires additional oversight. That means developers can move through longer coding tasks without repeatedly confirming commands or file changes.
Anthropic says its testing suggests this approach may actually be safer than relying on developers to manually review every permission request. In a study involving 1,053 paid Claude Code users, auto mode reportedly detected 89% of harmful actions, compared with 13.6% when users were responsible for reviewing prompts themselves.
One reason may be permission fatigue. Anthropic says users approved 97% of Claude Code permission prompts during manual review. When approval becomes routine, the permission dialog can stop functioning as a meaningful safety check and turn into another button developers click almost automatically.
Claude Code Head Boris Cherny also said in a post on X that his team has been using auto mode exclusively for months and that he could not imagine returning to the old permission-heavy workflow.
The move does not mean Claude Code is being given unrestricted access. Anthropic says it has been adding more safety controls around the autonomous workflow, including prompt injection screening and customizable hard-deny rules. These controls can block specific classes of actions, including attempts to move sensitive data outside an approved environment. More details are available in Anthropic’s official announcement.
The change reflects a broader shift happening across AI coding tools. Early coding assistants mainly suggested code while developers remained responsible for nearly every action. Newer coding agents are increasingly expected to inspect repositories, modify files, execute commands, test their work, and complete multi-step tasks with far less intervention.
For product and engineering teams, that changes the role of human oversight. The question is becoming less about whether a developer should approve every individual action and more about how teams define the boundaries within which an AI agent is allowed to operate.
Anthropic’s decision to make auto mode the default is another step in that direction. Claude Code is moving from an assistant that waits for frequent permission toward an agent that is expected to keep working unless it encounters a meaningful risk.
