Claude Code Introduces Direct Computer Control for Developers

Claude Code now enables direct control of computers, allowing developers to automate coding tasks and debugging seamlessly.

Claude Takes Off

Claude Code has officially launched its “computer usage” capability, allowing direct control of the CLI to write code, interact with UIs, and debug issues. This feature liberates developers from manual tasks with a simple command to activate “autopilot” mode.

New Capabilities of Claude Code

With the new feature, Claude is no longer just a chat AI hidden behind a dialog box. It can now operate directly within the CLI, taking over tasks like a real programmer. Claude Code can autonomously complete the development, debugging, and testing loop, executing tasks from writing code to compiling, launching applications, and automatically selecting tests.

If a program crashes, Claude can search for bugs, fix them, and validate the solutions. This has caused quite a stir among users, with some claiming that this marks the moment Claude officially replaces software engineers.

Currently, this functionality is available in a “research preview” for Pro and Max users, limited to macOS.

The Power of Claude Code

Claude Code can read your entire codebase, write code, modify files, and execute commands. However, its capabilities were previously confined to the terminal and text-based environments. Now, Claude can take control of a Mac, manipulating browsers, mice, keyboards, and screens to complete tasks.

Simply inputting /mcp in the terminal activates Claude’s “autopilot” mode. After integrating computer usage capabilities, Claude Code can perform the following operations:

  • Cross-application interaction: Open various installed apps and interact with UI elements.
  • End-to-end loops: Complete a full task cycle with a single command: write code -> compile -> launch app -> automate UI clicks -> find bugs -> fix code -> validate.
  • Ignore tool boundaries: Operate directly on local SwiftUI applications, Electron projects, or graphical tools without a CLI.

The core breakthrough of this update is the ability for “autonomous debugging” in complex environments. Previously, when code encountered issues, developers had to manually screenshot or copy error messages to report to AI, which was time-consuming. Now, Claude can directly view the program interface and simulate user actions to identify visual or logical flaws, significantly reducing the cost of switching between different tools.

A Complete Development Cycle

This update allows developers to achieve a complete development cycle without hands-on involvement. However, without sufficient quotas, developers can only watch their coding dreams fade away.

Quota Issues Arise

Today, Claude faced significant issues as global developers hit the “quota wall” of Claude Code. Even those who paid $200 for the “Max premium user” plan received quota warnings before fully utilizing the service. This sudden throttling has led to widespread complaints online about “Claude quota shortages.”

Serious Bugs Discovered

A Reddit user conducted reverse engineering on Claude’s binary files and uncovered shocking truths: two serious bugs in the system’s underlying architecture led to cache failures. When cache fails, token consumption can surge by 10-20 times, severely impacting user quotas. The first bug involves a string replacement issue in the Bun runtime environment, causing frequent cache failures. A temporary solution is to use npx @anthropic-ai/claude-code to run Claude. The second bug causes a 100% cache crash when using the --resume command to restore sessions.

Hidden Skills of Claude Code

Boris Cherny, the creator of Claude Code, shared 15 underrated “hidden skills”:

  1. Compiler in Your Pocket: Claude Code has a mobile app that allows coding on the go.
  2. Instant Cross-device Movement: Use /teleport to pull cloud sessions to local terminals.
  3. Ultimate Automation: Set Claude to automatically execute tasks using /loop and /schedule commands.
  4. Lifecycle Control with Hooks: Automatically trigger logic at specific times.
  5. Remote Commander: Use Cowork Dispatch for remote control of non-programming tasks.
  6. Frontend Development Eyes: A Chrome extension allows Claude to see outputs in real-time.
  7. Automatic Server Testing: Built-in capabilities for running web servers and testing.
  8. Session Forking: Create branch conversations with /branch to experiment without disrupting progress.
  9. Side Queries: Use /btw to ask questions without interrupting main tasks.
  10. Massive Parallelism: Run multiple Claude instances using claude -w.
  11. Batch Processing: Use /batch for large-scale code migrations.
  12. Speed Boost: Use --bare mode to skip configuration scans and speed up startup.
  13. Cross-repo Collaboration: Use --add-dir to access other folders during startup.
  14. Custom Agents: Define specialized agents for specific tasks.
  15. Voice Commands: Most of Boris’s code is dictated, using /voice in the CLI.

Mastering these skills can transform Claude Code from an assistant into a 24/7 engineering team.

Was this helpful?

Likes and saves are stored in your browser on this device only (local storage) and are not uploaded to our servers.

Comments

Discussion is powered by Giscus (GitHub Discussions). Add repo, repoID, category, and categoryID under [params.comments.giscus] in hugo.toml using the values from the Giscus setup tool.