Maya: Hey builders! Here's what's on the radar today on Claude Code Review. Cloud sessions run jobs with your laptop closed, Code Review hits general availability, Anthropic details how Claude writes ninety percent of Claude Code, faster replies land, and fork subagents expand. I'm Maya.

James: And I'm James. Let's get into it.

Maya: So Claude Code cloud sessions let you hand off a long task, close your laptop, and come back to finished work. Claude clones your repo from GitHub into a virtual machine and works there on its own branch.

James: That's the key part, Maya. Because the work lives on that remote machine, your laptop can sleep. A video from October fourteenth shows the loop: shut the lid, open the Claude app on your phone, and answer Claude's waiting question.

Maya: Right, and you start sessions from desktop, browser, terminal, phone or Slack. Each gets its own branch, so you run several at once without juggling work trees. Good for a big migration, or an idea that arrives away from your computer.

James: Fair caveat, though. One builder ran a Laravel app in cloud sessions on a two hundred fifty dollar credit and still said they prefer their laptop. And whether web sessions run slower or use more tokens, the sources raise but don't answer.

Maya: On to Code Review, James. Anthropic moved it out of research preview. It is now generally available on Claude Team and Enterprise, a finished, supported feature. Admins switch it on, select repositories, and Claude reviews new pull requests automatically.

James: And the speed jumped, Maya. Most reviews now finish in six to ten minutes, about three times faster than the roughly twenty they took in preview. Findings are concise: the problem, what triggers it, and the fix in a few sentences.

Maya: The main new control is depth. Admins pick Quick, Standard, or Deep per repository, so a routine codebase gets a quick pass while critical ones get a deep review. They enable it in settings on claude.ai and install the GitHub App; developers then do nothing.

James: Exactly right, and Anthropic runs Code Review on its own pull requests. On cost, eligible Enterprise organizations can claim at least ten thousand dollars in Code Review credits by January fifteenth, twenty twenty-seven. Credits expire thirty days after you claim them.

Maya: Now the backstory, James. Anthropic says Claude now writes about ninety percent of the code in Claude Code, and it credits reworked code review. From June to September, merged pull requests more than doubled.

James: And the share of changes reverted fell by more than forty percent, Maya, meaning fewer accepted changes had to be rolled back. Their write-up says the engineer's role was very often to tell Claude no.

Maya: Right, because agents will sometimes break core system constraints to satisfy a pattern that should never have been allowed. They say this was particularly common in their migration, where systems read the commit log instead of asking the review tool.

James: And they lean on adversarial review, one AI pass set up to challenge another's work. They call it incredibly effective but myopic, so skilled engineers had to steer Claude away from over-engineering for hypothetical concerns.

Maya: Two speed changes now, James. On October sixteenth, ClaudeDevs said Claude Code gzips the conversation it sends on every turn, so the same data travels as fewer bytes and reaches the model faster.

James: Exactly, and the second one is cloud sessions starting up as soon as you start typing. The team says you may have noticed Claude starts responding sooner. Small changes, but they shave real time off each back-and-forth.

Maya: Last one, James. Claude Code two point one point three oh two turns on fork subagents in more places. A fork subagent inherits the parent conversation instead of starting with empty context.

James: Right, it was already on in the terminal, now also in claude dash p, the Agent SDK, desktop, and VS Code. The risk sits with automated jobs: if a CI script depended on helpers starting blank, the output can now differ.

Maya: Good flag, James. A fork starts with the parent's context, which can mean more input tokens up front, and the changelog doesn't quantify the cost. To keep the old behavior, set the fork subagent environment variable to zero.

Maya: That's your Claude Code Review daily brief. Check claudecodereview.ai for the full stories. Keep building.