AI coding tools like Claude Code have created a problem nobody expected. Engineers are shipping code faster than ever but all those pull requests still need to be reviewed by a human. At Anthropic alone, code output per engineer grew 200% in the last year. The review queue became the bottleneck. Today Anthropic launched Code Review inside Claude Code to fix that exact problem.

When a pull request opens, Code Review dispatches a team of AI agents to review it in parallel. Some agents scan for bugs and risky patterns. Others verify the findings to filter out false positives. A final pass ranks every issue by severity and impact. The result lands directly on the pull request as a single high-signal summary plus inline comments for specific bugs. The whole process takes about 20 minutes and costs between $15 and $25 per review depending on size.

Lets Look At A Quick Demo From The Official Claude Drop Below 👇

Part 1

Part 2

Part 3

Part 4

Look at what just happened in those four screenshots. A developer typed one sentence. Claude went away, completed every task on the list, and came back ready to create the PR. Then during the review it caught something that would have kept most engineers up at night. A critical security flaw that let any authenticated user access anyone else's account tokens just by guessing session IDs. CVSS 9.1. Full account takeover. Already in the codebase. Already about to ship. Claude caught it, explained exactly why it was dangerous, told the developer precisely how to fix it, and the commit was verified. That bug does not make it to production. That is the whole point.

Before deploying Code Review internally, Anthropic engineers received substantive feedback on about 16% of pull requests. After turning it on, that number jumped to 54%. For large pull requests over 1,000 lines, 84% of reviews surface real findings. Engineers disagreed with less than 1% of flagged issues. That is not a beta quality tool. That is a tool they trust on almost every PR they ship.

Code Review does not flag style issues or formatting nitpicks. It focuses exclusively on logic errors because those are the bugs that actually break things. It scales dynamically too. Larger or riskier pull requests get more agents and deeper analysis. Smaller updates get a lighter pass. The system knows the difference and adjusts automatically.

That dynamic scaling is also why the pricing works the way it does. A small focused PR costs around $15. A large complex one with hundreds of changed files runs closer to $25. You are essentially paying for more agents and deeper reasoning on the reviews that actually need it. For most engineering teams that is still cheaper than the hour of senior developer time a thorough manual review actually costs. The math is not even close.

Today Microsoft launched Copilot Cowork, built in direct collaboration with Anthropic and powered by Claude Cowork technology. It is part of Wave 3 of Microsoft 365 Copilot and it does something no previous version of Copilot could. It does not just answer questions. It executes work across your entire Microsoft 365 environment autonomously.

Here is what that looks like in practice. You tell Copilot Cowork to prepare for an important client meeting. It pulls your emails, reviews past meeting notes, checks your calendar, builds a briefing document, creates a PowerPoint deck, and schedules prep time. All without you opening a single app. You come back and everything is ready. You can approve, adjust, or redirect at any step.

The original Claude Cowork runs locally on your desktop. Copilot Cowork runs in the cloud inside Microsoft 365 and has access to your entire work graph. Every email thread, every Teams conversation, every SharePoint file, every Excel workbook, and every meeting in your history. That is a completely different level of context and a completely different category of tool.

Pricing lands at $99 per user per month through the new Microsoft 365 E7 bundle, which packages Copilot, identity management, and the new Agent 365 platform for managing AI agents across an organization. Copilot Cowork is in limited research preview now and rolling out through Microsoft's Frontier program in late March.

The irony of today is hard to miss. The U.S. government banned Anthropic last week. Today Microsoft, the world's largest software company, put Claude inside the productivity suite used by 90% of the Fortune 500. The market has spoken.

Think about what happened today. Anthropic launched a marketplace that turns Claude into an enterprise ecosystem. They shipped Code Review that makes Claude responsible for auditing its own code. And through Microsoft they put Claude inside the daily workflow of hundreds of millions of office workers around the world. All of that happened while Anthropic is actively fighting the U.S. government in court. The Pentagon tried to cut them off from their biggest customer. Instead Anthropic went and found hundreds of millions of new ones. That is not a company on defense. That is a company that decided the best response to getting banned was to become impossible to ignore.

Stay building. 🤖