article — General

Learning Resource

About this Resource

Look at the biggest Claude Code updates, from multi-agent collaboration and Auto Mode to better code reviews, design tools, browser access, and iOS development.<h2> </h2><h2>Claude Code Gets a Major Upgrade: The Biggest New Features</h2>Claude Code has received a major set of upgrades that make it more capable of handling complex coding tasks with less manual supervision. The updates focus on helping AI agents work together, improving automation, making development faster, and giving Claude better tools for building and testing applications. The biggest change is that Claude Code is becoming more than a coding assistant. It is increasingly designed to act as an AI coding agent that can plan, use tools, work across multiple tasks, and continue working with less human intervention. <h2>AI Agents Can Now Communicate With Each Other</h2>One of the most useful additions is cross-session messaging. Previously, separate Claude Code sessions operated mostly independently. Now, different sessions can communicate with one another and share important information. For example, one Claude session could work on a website's frontend while another works on the payments system. If an API name changes in the payments session, the relevant information can be sent directly to the frontend session instead of manually explaining the change again. This also enables more advanced workflows where multiple AI agents work on different parts of a project at the same time. <h2>Auto Mode Reduces Interruptions</h2>Claude Code is also making Auto Mode the default permission mode for Pro, Max, and Team plans. Instead of repeatedly stopping to ask for permission before performing actions, Claude can automatically decide when certain actions should be executed. This is particularly useful for long coding tasks where constant interruptions can slow everything down. Anthropic's testing, as presented in the source material, found that Auto Mode blocked significantly more dangerous commands than the human participants in its study. Anthropic also reported that teams using Auto Mode shipped roughly 25% more pull requests, although these figures come from Anthropic's own testing and should be viewed in that context. Users can still change the permission mode, while existing custom or organization-managed defaults are not automatically replaced. <h2>A New Concise Mode Saves Tokens</h2>Claude Code now includes a Concise Mode that tells Claude to keep its responses shorter and focus on the result. This matters because long explanations can consume additional tokens during extended coding sessions. With concise mode enabled, Claude can provide the important information first while leaving the user the option to request more detail when needed. For developers working through many tasks, this can make interactions with Claude faster and more efficient. <h2>Faster Performance and More Usage</h2>Several improvements target Claude Code's underlying performance. The desktop application now launches roughly twice as fast compared with the previous month, while the CLI has also received GPU-usage improvements. Claude Code's weekly usage limits have also received a temporary 50% increase, with Anthropic indicating that it hopes to make the increase permanent. Capacity remains an issue because of continued demand. <h2>Claude Can Now Help Design the Interface</h2>A new Design Skill is available as a research preview. Using the /design command, users can create editable UI artboards, make visual changes, and then have Claude implement the resulting design in an application. This creates a tighter connection between design and coding. Instead of describing a design entirely through text, developers can work with a visual representation and then turn that design into working code. <h2>Building iOS Apps Gets Easier</h2>Claude Code's desktop app now includes an integrated iOS Simulator on macOS. Claude can build and launch an iOS application while the simulator appears alongside the conversation. It can interact with the application, move through screens, and help verify changes. This brings building, testing, debugging, and iteration closer together, reducing the need to constantly switch between different applications. The feature is available in public beta for Pro, Max, and Team plans and requires Xcode with the iOS platform installed.<h2> </h2><h2>Better Code Reviews</h2>Claude Code has also improved its code review system. Different effort levels now determine how deeply Claude examines a change:<ul><li>Low effort: A quick review designed for speed. </li><li>Medium effort: Multiple passes with additional context and verification. </li><li>High effort: Uses additional agents to find and verify potential problems. </li><li>Extra-high effort: Looks for issues that could appear outside the code directly changed. </li><li>Ultra: Uses multiple reviewers and agents to independently investigate findings and reduce false positives.</li></ul> The key idea is simple, developers can choose between faster reviews and deeper analysis, depending on how important the code change is. <h2>Claude Artifacts Can Use MCP Connectors</h2>Claude's artifacts can now work with MCP connectors, allowing generated dashboards, tools, and interactive applications to access live information and perform actions. This makes artifacts more useful than simple static interfaces. They can become functional applications that interact with connected services. MCP access is currently disabled for publicly shared artifacts, while the feature is available across paid tiers. <h2>Claude Code Gets Its Own Browser</h2>The Claude Code desktop app now includes a built-in browser. Claude can use it to open websites, documentation, dashboards, and other resources, read information, and interact with pages while working on a task. This means developers don't necessarily have to copy information from a browser into Claude manually. Claude can access relevant web-based context directly within its working environment. The browser is sandboxed and configurable, including controls around whether sessions can log into websites using stored credentials. <h2>Better Visual Feedback</h2>Claude Code also supports a more visual way of communicating changes. Users can highlight parts of an interface, draw on areas, add text, or point out specific elements they want changed. Claude can then use that visual feedback to understand what needs to be modified. For UI development, this can make the feedback loop much simpler than trying to describe every visual problem using text alone. <h2>What These Updates Mean</h2>The biggest shift is that Claude Code is becoming a more autonomous development environment. Instead of simply asking an AI to write a piece of code, developers can increasingly give Claude larger objectives and let it handle multiple parts of the development process: coordinating agents, researching information, writing code, reviewing changes, testing applications, interacting with websites, and responding to visual feedback. Combined with powerful models, these improvements make Claude Code better suited to longer and more complicated development workflows. The important takeaway is that AI coding tools are moving toward a model where you describe what you want to build, while the AI handles more of the work required to turn that idea into a working application.