Claude Code, Anthropic's desktop application for AI-assisted development and knowledge work, has undergone significant updates that make it far more practical for everyday users than it was just months ago. The app now includes multi-pane sessions that let you run multiple AI agents simultaneously, a built-in browser for real-time visual feedback, artifact creation with live data connections, and scheduled routines for recurring tasks. These features suggest Anthropic is positioning Claude Code not just as a developer tool, but as a general-purpose AI workspace.

What New Features Does Claude Code Actually Offer?

The latest Claude Code updates address a common misconception: that the tool is only for programmers. The desktop app has evolved significantly since earlier versions that users described as sluggish and unintuitive. Today, it includes several features designed for broader knowledge work and creative tasks.

• Multi-Pane Sessions: You can drag conversations from the sidebar into the main window and arrange them side-by-side or stacked, allowing multiple AI agents to work on different tasks simultaneously without switching between windows.
• Built-In Browser with Visual Feedback: The app includes a browser that displays live previews of changes as agents make them, and you can annotate or draw on webpage elements to give Claude visual instructions instead of describing problems verbally.
• Artifacts with MCP Connectors: Artifacts are shareable HTML dashboards and tools hosted on Anthropic's servers; the latest update lets them connect to MCP (Model Context Protocol) connectors, enabling real-time data pulls and actions from external apps like Substack or Google Analytics.
• Scheduled Routines: You can set up recurring tasks that Claude runs on a daily or weekly schedule, such as checking emails and generating to-do lists or pulling weekly analytics reports.

How Should You Actually Use Multi-Agent Sessions?

While the multi-pane view makes it possible to run many agents at once, early testers suggest restraint is wise. Running too many parallel sessions creates confusion rather than productivity. Most experienced users keep between two and four agents active at a time; beyond that, tracking what each agent is doing becomes difficult.

The app now displays activity status clearly, showing which sessions are still working and which are waiting for user input. For people managing multiple projects, grouping agent sessions by project in the sidebar provides a clear overview of progress across different folders and tasks.

What Permission Modes Does Claude Code Offer?

Claude Code includes three permission levels that control how much autonomy the AI agent has before asking for approval. Understanding these modes is critical for safe and effective use.

• Manual Mode: The most restrictive option; Claude asks for permission before taking most actions, making it ideal for users new to AI agents.
• Auto Mode: Claude can execute much of the work independently but pauses when facing potentially destructive changes, such as deploying website updates to production or removing important files. This mode balances autonomy with safety guardrails.
• Bypass Permissions: The most permissive mode, where Claude continues working without asking for confirmation on standard actions, though this carries higher risk if the agent makes unintended changes.

Anthropic appears to have designed Auto Mode with intentional guardrails. Users report that Claude will stop before deploying live website changes or making destructive edits to code, even when working in Auto Mode. This suggests the company is prioritizing safety over pure convenience.

How Can You Use the Built-In Browser for Real Work?

The integrated browser is one of the most practical additions to Claude Code. Instead of switching between the app and a separate browser window, you can see live previews of changes as Claude makes them. This is particularly useful for web design, content updates, and visual debugging.

The browser also supports visual annotation. You can select an element on a webpage, draw or highlight the area you want changed, take a screenshot, and add it to the chat. Claude then understands exactly what you mean without requiring verbose verbal descriptions. The browser can also maintain login sessions to services like Substack, X, or LinkedIn, enabling agents to help with tasks like scheduling posts or conducting research.

However, caution is warranted for any task involving publishing, purchasing, or sending messages. While seeing the agent's browser actions in real time is more transparent than handing off a task blindly, users should still monitor sensitive operations closely.

What Are Artifacts and How Do They Work Now?

Artifacts are interactive HTML tools and dashboards hosted on Anthropic's cloud infrastructure, making them shareable with others via a link. Recent updates allow artifacts to connect to MCP connectors, which means they can fetch live data from external services and take actions on behalf of each viewer.

This makes artifacts useful for real-time dashboards, reporting tools, and presentation-style outputs. Users report creating operational dashboards that pull data from Substack, Google Analytics reports that update automatically, and cohort presentations that are easy to share with teams. The ability to edit shared artifacts means multiple people can collaborate on the same tool without creating duplicate versions.

When Should You Use Scheduled Routines?

Routines are Claude Code's answer to recurring tasks. If you have work that needs to happen on a regular schedule, whether daily or weekly, a Routine can automate it. Examples include checking emails and generating a prioritized to-do list, pulling weekly analytics reports, or monitoring specific metrics.

The feature is still in early stages, but it represents a shift toward AI agents handling ongoing work rather than one-off tasks. This aligns with broader industry trends toward agentic AI, where systems take initiative on scheduled or event-driven work rather than waiting for explicit user requests.

The updates to Claude Code suggest Anthropic is moving beyond positioning Claude as a chatbot and toward building it as an operating system for AI-assisted work. The combination of multi-agent sessions, visual feedback loops, real-time data connections, and scheduled automation creates a platform where knowledge workers and developers can delegate entire workflows to AI, not just individual questions or coding tasks.