Claude Code launches experimental Code Mode: multiple tools in a single call, long tasks can run in the background
Beating AI News Flash: Anthropic has added an experimental Code Mode to Claude Code, now available in v2.1.323 released on November 12, and currently requires manual activation.
Once enabled, Claude can first write JavaScript scripts, then use the scripts to call multiple tools sequentially and in parallel, and can also pre-process the data returned by tools, passing only the necessary results to the model, thereby reducing model requests and Token consumption.
The new mode also supports long-running tasks in the background. Claude can continue answering users while executing commands. Official documentation shows that background scripts can run for up to 2 hours, but automatic authorization modes such as Auto are required.
Code Mode was proposed by Cloudflare in September 2025, and Anthropic also publicly disclosed a similar solution in November of the same year. This year, open-source Agent frameworks such as DeepSeek Harness, Pi, and OpenCode 2.0 have successively added support.
Anthropic has only now integrated it into Claude Code, and it remains in the experimental stage and requires manual activation.