How Claude Code's team checks Claude's work
Claude blog · Addy Osmani · Nov 16
Tools
Plus Claude Sheets against Excel and Google Sheets
Written with AI assistance, reviewed by the ACN desk.
Why it matters
Coding agents are getting more autonomy and tighter guardrails at once: Claude Code asks before more commands run, while Code Mode lets it do more in one step.
Anthropic gave Claude Code more power and more brakes in the same week. Sandboxed commands now ask before they run, and a new Code Mode lets Claude write programs to drive its own tools. It is the tension every coding agent faces: more autonomy, and tighter control. Claude also gained three new work tools, in a big week for Anthropic.
On Friday, November 13, Claude Code 2.1.325 stopped treating the sandbox as permission. Sandboxed Bash commands now follow the normal permission flow and ask before running; sandbox.autoAllowBashIfSandboxed: true keeps the old behavior. It was the fourth release in a week to tighten permissions. The release notes give no reason. The change comes after a year of prompt-injection attacks on coding agents. In April, researchers hijacked Claude Code’s security review agent, Gemini CLI and GitHub Copilot through GitHub comments, SecurityWeek reported, and Anthropic rated its case critical. When Anthropic introduced the sandbox in 2025, the aim was fewer permission prompts.
A day earlier, version 2.1.323 added Code Mode, “an experimental opt-in in which Claude writes JavaScript to call its tools.” The idea is spreading across the industry. Cloudflare named it in 2025, arguing that models write code better than they make tool calls. Anthropic’s engineers reported one task falling from 150,000 tokens to 2,000 when the agent wrote code instead. For developers, the trade is agents that do more in one step, and more moments where they stop and ask.
Also on Friday, Anthropic updated Projects in Claude Code, its public beta on Pro and Max plans. In a project, Claude coordinates tasks across parallel threads. The coordinator is now “much more proactive,” Anthropic says: it answers questions in the project chat and reports on ongoing threads. A thread can be shared as a read-only link with anyone signed in to Claude.
On Tuesday, November 10, Anthropic launched Claude Whiteboards and Claude Sheets in beta for all paid plans. Whiteboards is a shared board for brainstorming with Claude; Sheets is a spreadsheet Claude builds and edits with you. Claude Meetings, in beta for Team and Enterprise plans, lets Claude join Google Meet calls as a visible participant and start work during the meeting. Anthropic says it saves a transcript, summary and screen descriptions, not audio. For developers, self-hosted Claude Managed Agents entered private beta.
The same week, Anthropic released Claude Fable 5.2 and Mythos 5.2 on Monday, November 9, at $8 per million input tokens and $40 per million output tokens. Fable 5.2 is now the default Fable model in Claude Code. On Tuesday, November 10, Anthropic launched the roadshow for its proposed IPO, at an expected $70 to $78 a share. Reuters reports the top of that range would value Anthropic at about $1.4 trillion fully diluted, or about $1 trillion on a non-diluted basis, and the offering would raise up to $64.4 billion. That would make it the second-largest IPO ever, behind SpaceX’s $75 billion in June, according to the Motley Fool’s list.
Coming up: Anthropic expects to set the IPO price on Wednesday, November 18, according to its IPO site. Claude Data, formerly Dashboards, turns on by default for Enterprise organizations on Friday, November 20, and Claude Sonnet 4.5 retires from the Claude API on Monday, November 30.
Sponsor
Spot an error? Email corrections@aicodingnews.si. Fixes are logged on the corrections page.
Sponsor
Sponsor
Claude blog · Addy Osmani · Nov 16
Version 2.1.323 · experimental, opt-in
Shares close up 17% after topping $2 trillion at the peak
Price weighed well above $80 · Nasdaq: ANT · trading Thursday, Nov 19
Sponsor
Sponsor