Claude Code makes auto mode the default for Pro, Max, and Team

New sessions skip per-call prompts, routing each tool call through a safety classifier. Plus Cursor's 3x-faster cloud agents and free GLM 5.2 on Vercel.

Nowline AUG 14 3:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • It stops asking today

    As of August 14, new Claude Code sessions on Pro, Max, and Team launch in auto mode by default: instead of approving each tool call, a classifier blocks only actions that are irreversible, destructive, or reach outside your environment. It falls back to manual approval after 3 consecutive or 20 total blocks in a session.

  • Why the prompts had to go

    Anthropic's data showed 97% of permission prompts were approved reflexively and 62% of users had already bypassed them. In a 1,053-tester study the classifier caught 89% of dangerous commands versus 13.6% for human reviewers, and teams on auto mode ship roughly 25% more pull requests.

  • Your opt-out switches

    Prefer manual? Shift+Tab (or the desktop mode dropdown) still flips modes mid-session, `disableAutoMode` removes auto mode entirely, and admins can pin `defaultMode` in managed settings. Enterprise, the API, Bedrock, Vertex, and Microsoft Foundry stay opt-in for now.

  • Set your deny rules before you trust it

    Before flipping it on for real work, run `claude auto-mode config` and `claude auto-mode critique`, prune broad allow-rules like `python:*`, and define `hard_deny` entries. Anthropic dropped classifier token charges for Pro/Max/Team but still recommends manual review for high-stakes infrastructure.

  • Elsewhere: Cursor cloud agents boot 3x faster

    Cursor's new Builds pre-bake your repo, dependencies, and setup scripts into a warm image, so cloud agents start 10x faster with 3x faster time to first token, and keep running on the last good build if a commit breaks the environment. Included free with Cloud Agents.

  • Elsewhere: GLM 5.2 free on Vercel's Gateway

    Z.ai's open-weights GLM 5.2, a 1M-token-context coding model, is free for Vercel 'eve' agents through August 27, and Gemini 3.7 Flash landed on the AI Gateway at a 50% discount behind one API key with automatic fallbacks.