Claude AI gets auto mode for tasks

Anthropic has updated its Claude AI system to include an auto mode that reduces the need for user approval at each step when performing tasks. This feature, added on October 22, will be deployed next week in "most" of its applications for new sessions of the Pro, Max, and Team packages.

Auto mode skips individual approvals

Users will no longer need to approve each action individually, as an automatic control reviews them. This system allows actions deemed safe to proceed while blocking riskier ones, and Claude searches for alternative ways to complete tasks. The company has not specified which applications are affected.

Claude Code already uses auto mode

Users of Claude Code, a programming assistant from Anthropic, are already using this setting. Since August 14, new Pro, Max, and Team sessions in Claude Code start in auto mode. Those who had previously chosen a different setting receive at most a single prompt offering to switch.

Auto mode reduces cost for users

Since August 7, Anthropic no longer deducts the cost of the filter from the usage quota of these subscribers in Claude Code. This control consumes a few more tokens with each action, but the change helps reduce expenses for users.

Second model checks for risky actions

A second model validates the commands on behalf of the developer. The auto mode sends each action of Claude through a classifier, a second AI model tasked with blocking irreversible, destructive, or actions that send data outside the work environment, such as code published on a public site.

Auto mode improves productivity

Anthropic justifies this choice with its own statistics: users approve 97% of permission requests, which shows, according to the company, that many click reflexively. In terms of productivity, users of the auto mode among its Team and Enterprise clients deliver about 25% more pull requests.

Other products gain more autonomy

Anthropic also gives more autonomy to its other products: with Claude Duet, its AI reads emails and takes initiatives on behalf of the user. Users can change the setting at any time using keyboard shortcuts or the desktop application menu.

Admins can control auto mode

Administrators can enforce a mode for the entire organization using the defaultMode parameter, or completely disable the auto mode with disableAutoMode. Anthropic acknowledges that its filter does not eliminate all risk and advises users to review Claude's actions themselves before any sensitive modification of a production infrastructure.