This is insane. Claude is about to become a real person. Last night, Claude Code was officially released. Now anyone can use it. It can write code directly in the CLI, build a UI, and fix bugs. It is like a fully automatic driving mode. Claude finally grew a pair of hands.
The speed of Claude’s evolution is getting faster and faster. It is becoming more and more powerful.
Just last ai video generator pornnight, Anthropic dropped another bomb.
Now everyone can use the official version of Claude Code.
This means that Claude is no longer just a chatbot in a browser. It is a real AI agent.
It has grown a pair of hands. It can now live directly in your CLI and take over your entire computer.
With Claude Code, you can become a real one-person girlfriend gpt army. You can write code, test it, and fix bugs all by yourself.
In other words, you just give Claude a prompt. It will write the code, test it, build the app, and do the full process automatically.
You do not need to write any code yourself. You do not need to fix any bugs yourself. You do not need to test anything yourself.
As soon as the news came out, the whole internet went crazy. People said that this was a historic moment. Claude had officially replaced programmers.
Right now, there is a “Research Preview” available. Pro and Max users can use it. It works on macOS.
Claude Code Grew Hands and Eyes
There is only one word left: insane.
In fact, Claude Code is already very powerful. It can understand your natural language requests. It can write code, read files, and run commands.
It is like having a senior engineer living in your terminal and text editor.
Here is an example. You open the terminal. You tell Claude what you want to build. It will write the code, build the UI, and give you the result.
From now on, Claude can directly take over your Mac. It can open the terminal, write code, run tests, and fix bugs.
The terminal becomes a real MCP client. Claude can drive it in automatic mode.
Here is the full process that Claude Code can execute automatically:
Write code -> Test -> Build App -> Auto-test UI -> Find Bug -> Fix Bug -> Re-verify
Every step is done by AI. The whole process does not need any human help.
When Claude Code runs into a conflict, it will stop and ask you for help. It will show you the conflict. It will wait for your decision.
Right now, Claude can directly see your project files. It can simulate the user experience. It can find logical holes in your code.
This “human-in-the-loop” mode has truly changed the way people work. It has greatly reduced the cost of switching between different logics.
ai furry porn
Here is a more powerful update. Claude Code now supports a complete full-process operation log.
Some people say that Claude Code is too powerful. Some people are worried. But most people are just excited.
Token Limits Hit: The Whole Internet Is Crying
But today, something unexpected happened on Claude’s official website.
Just one day after the release, developers around the world crashed into Claude Code’s “paywall.”
Even some users who paid for the “Max” plan, which costs 200 dollars, hit the limit. They ran out of tokens.
This wave of crashes directly pushed Claude to the top of the global trending list. Claude was trending everywhere.
Some people joked that in less than one year, Claude had already “graduated.”
Click to view the full image
Some people said that their local version of Claude Code also had problems. They could not use it.
But the reason is still the same. It is still the token limit.
Three Major Bugs: Token Costs Went Up by 20 Times
A Reddit user captured something shocking.
He used a man-in-the-middle attack to intercept Claude’s network requests and engineering logs. He found three shocking bugs:
The system-level repeated retry bug directly caused the account to become invalid.
You need to know that one failure can cost 10 to 20 times more tokens than normal. This is basically killing the user’s wallet.
Some people said that the API is still usable. But only for simple tasks. For complex tasks, the bug is still there.
Another bug is that Bun’s temporary file character replacement bug caused Claude’s CLI to crash. The auto-retry mechanism failed.
Right now, no one knows when this will be fixed.
The second bug is that when using the browser tool, the success rate is only 100 percent. Wait, that means it always works?
No. It means the “success” rate is a bug. The browser tool always reports success, even when it fails.
In the discussion, some people said that Anthropic should fix the bug or give users a refund.
Right now, GitHub has a large number of bug reports. People have confirmed the leak of the retry mechanism. The “token burning” incident has been recorded. It may take some time to fix.
Claude Code Hidden Skills: 15 Secret Weapons
Today, Claude Code expert Boris Cherny released 15 hidden low-key “secret weapons.”
Secret 1: Build iOS Apps Directly on the Phone
Many people do not know that Claude Code can build iOS apps directly on the phone. Boris tested it on an iOS app and wrote the code directly.
He used a remote build team to wait for the code to compile. Then he directly submitted the code through the Code tab on his phone. It was like a real “submit PR” process. It was amazing.
Secret 2: Device-to-Device Handoff
Imagine this. You are working on a project at the company. You start a Session on your laptop. Then you close your laptop and go home. You only bring your phone.
You use the /teleport command to move the session to your phone. Then you use the /remote-control command to control the session on your laptop from your phone. It is like magic.
Secret 3: Auto-run and Dynamic Mode /loop and /schedule
Boris’s favorite feature is the ability to let Claude run automatically for a long time, up to one week.
Secret 4: Custom Hooks Trigger Auto-run
Use Hooks to set up logic. Trigger auto-run at specific times.
Secret 5: Remote Command Center Cowork Dispatch
When you are not in front of your computer and need to handle Slack messages, file transfers, or MCP tasks, you can use Dispatch.
This is like the safe remote control of Claude Desktop. It can handle non-urgent tasks for you.
Secret 6: Front-end Testing Eye Chrome Extension
Boris’s most wanted feature is to let Claude see the web page in real time through the Chrome extension. Claude can see the actual rendering effect and element relationships.
Without this feature, Claude is like a blind person. It can only guess what the front-end page looks like based on the code.
Secret 7: Auto-test and Auto-deploy
The Claude Desktop App already supports auto-testing of web pages. It can automatically run tests and find bugs in the project.
This feature is much more useful in the CLI than in the web version.
Secret 8: Temporary Mode /tmp and Fork Session
Here is a bold idea. What if you mess up the current progress?
You can instantly create a temporary branch in the CLI. It has the same effect as a fork.
Secret 9: Side Task Mode /btw
When Claude is writing code and suddenly remembers another task, what should it do?
Use the /btw command to create a temporary sub-task. Claude will handle it in the background. Then it will come back to the main task.
Secret 10: Parallel Mode Git Worktrees
Boris’s secret weapon is to run dozens of Claude agents at the same time.
He uses Git worktrees to create multiple working copies in the same repo. Each copy can run independently.
Secret 11: Text-to-Image Batch Mode /batch
When you need to generate a large number of images, you no longer need to talk to the AI one by one. You can automatically distribute thousands of tasks to different agents and run them at the same time.
Secret 12: Speed Up by 10 Times –bare Mode
By default, Claude scans all files in the project in real time.
In non-interactive mode, use the –bare mode to skip these scans. The SDK speed can be increased by 10 times.
Secret 13: Custom Workspace Boundary –add-dir
When you need to collaborate across repos, you can use –add-dir to add other directories. Claude will only access files within the allowed range.
Secret 14: Custom Agent Mode –agent
Create a dedicated agent for each task. For example, one agent for security review. One agent for code review.
They can run in parallel and improve efficiency.
Secret 15: Voice Input Mode /voice
Boris said that in his actual use, most of the input is “spoken” rather than typed.
In the CLI, press the space bar and hold it. Speak your command. Release the space bar. Claude will execute it.
With these secret weapons, Claude Code is no longer just a tool. It has become a 24-hour working team.
