Anthropic、「Claude Code」に新機能「auto mode」を搭載–一部プランから – ZDNET Japan https://www. yayafa.com/2773376/
agents anthropic claude
| Source: Mastodon | Original article
Anthropic has rolled out “Auto Mode” for its Claude Code developer assistant, making the feature available to customers on the Team research‑preview plan and the Enterprise tier. The new mode lets Claude Code execute code, install packages and manipulate files on a user’s machine without prompting for each permission, a step up from the earlier “‑dangerously‑skip‑permissions” flag. Auto Mode is currently compatible with the Claude 3.6 Sonnet and Claude 3.6 Opus models, and Anthropic describes it as a safety‑enhanced bridge between fully manual approval and unrestricted execution.
The addition matters because it narrows the gap between Anthropic’s offering and rival AI‑powered coding tools such as GitHub Copilot, OpenAI’s Code Interpreter and Microsoft’s recent AI extensions for Visual Studio. By automating routine scripting and debugging tasks, Auto Mode promises to cut development cycles and lower the cognitive load on engineers, especially in large‑scale codebases where repetitive refactoring is common. At the same time, Anthropic stresses that the feature does not eliminate risk; it merely raises the safety bar while still requiring developers to monitor outputs and retain ultimate control.
What to watch next is how quickly the feature moves out of research preview into a broader rollout. Enterprise adopters will likely test the balance between speed and security, and any incidents of unintended code changes could shape Anthropic’s safety roadmap. Competitors are expected to respond with their own autonomous execution modes, potentially sparking a rapid arms race in developer‑centric AI. Regulators and industry groups are also beginning to scrutinise “self‑acting” AI agents, so policy developments could influence how Auto Mode is packaged and disclosed in the coming months.
Sources
Back to AIPULSEN