Netizens' experience of using Claude Code

I've been using Claude Code for a while recently, and I must say, this tool is ruthlessly efficient at cleaning up my old messy codebase, like a dollar-driven paper shredder. Its strength lies in the fact that it can accomplish astonishingly complex tasks just through conversation.

You don't even have to select the context yourself; as long as you open your heart (and wallet), Claude Code will automatically take control of the entire situation. However, it is considerate too, asking every few seconds if it is allowed to execute some basic read-only commands—commands that you can safely entrust to anyone, even North Korean hackers.

But you need to keep an eye on it because its momentum in advancing progress is very strong. As long as the funding authorization remains in place, it will continue to track bug fixes until the code is fixed and deployed to the production environment, and even start analyzing user logs to see how it performs.

Although Claude Code's interface is indeed quite outdated and lacks multi-modal support, making collaboration with other tools somewhat troublesome, these issues are not important. Because despite its outdated appearance, it makes other tools—whether Cursor, Windsurf, Augment, or our own tools, and even Copilot—seem instantly obsolete.

I know Claude Code is still in the experimental stage, and we haven't fully figured out its limits yet, but at least from my current experience, it may be the most revolutionary leap since the advent of programming assistants.

In short, Anthropic not only has the best code model in the industry, but also seems to know better than anyone else how to use them.

Watching Anthropic continuously release the strongest models, the best chat interfaces, precise predictions by the CEO about the industry, and now Claude Code, I begin to suspect that Anthropic might really be the only company on Earth that knows what it's doing.

Claude Code (CC)It is a brand-new experimental programming assistant recently launched by Anthropic, which is completely different from other products on the market. Surprisingly, it can only run through a terminal (terminal), such as common xterm or bash shell, or even six distinctive yet flawed WSL shells.

There was once a tool called Devin, which was quickly forgotten after being ridiculed upon release, but Claude Code is different—it is truly an AI software engineer in every sense. More surprisingly, it seems to have completely abandoned the popular RAG technology, allowing Claude itself to figure everything out.

After experiencing it for a period of time, I have to admit that I can no longer live without Claude Code. It's like a programmer version of "Assassin's Creed," where players don't even need to move, just issue orders and enjoy the results, leaving the rest to the AI to complete. Claude Code will continuously handle tasks automatically and constantly tackle complex coding problems: "snap, snap, snap" solving problems quickly and efficiently.

Imagine issuing a command to it: "My project has accumulated years of technical debt, help me write a perfect set of test cases, smart enough, meaningful, and consistent with our existing testing patterns." Then, just throw a few thousand dollars into its bottomless fund pool, and Claude Code will quietly work on a branch for one or two weeks, diligently helping you finish all the work.

I dare say that all future programming assistants will benchmark against Claude Code, following its steps to improve their products. Now, all competitors are scrambling to imitate, because, facing the old question you often ask: yes, Claude Code is that powerful.

A new era of competition has arrived!

When starting a new project, please make sure to use/initcommand.

How to use Claude Code well?

There are some practical experiences worth sharing when using Claude Code:

1. Clearly tell Claude Code what you want

When you plan to perform large-scale refactoring or add complex features,make sure Claude Code provides a detailed plan firstbefore starting specific implementation.

You can use the following expressions to stimulate its deep thinking:

• Think hard (deep thinking)
• Think deep (deep thinking)
• Think longer (long-term thinking)

After Claude Code provides the implementation plan:

• If you're satisfied, you can proceed directly with the changes;
• If you have questions, then communicate further with it.

2. Instructions must be clear and specific

When giving instructions to Claude Code, be sure to clearly specify the files and content you want to modify, for example:

• "Claude, please help me adjust the size of the chat input box in theuser/page.tsxfile, making it larger."
• "Please optimize thedata_processing.pyfile."
• "Help me modify theinvoice.pyfunction insidecalculate_totals."

3. Planning is essential before major changes

When preparing for large-scale code refactoring or adding complex features, always let Claude Code provide an implementation plan first, evaluate it, and then act.

Other best practices for using Claude Code:

• Regularly use/compactcommand
  Use/compactcommand more during development to help Claude Code work efficiently and avoid confusion or misunderstandings that may arise in long-term projects.

• Clearly specify the files to be modified
  Each time you give Claude Code an instruction, clearly point out which files you want to modify and explain their relationship with other components.

• Run tests frequently
  Claude Code excels in testing and validating new features, ensuring the reliability and accuracy of code functions through continuous and thorough testing.

• Periodically review the entire project
  Regularly ask Claude Code to review the entire project, assess the consistency of the code logic, identify redundant or inefficient code, and ensure the overall architecture is healthy and efficient.