Can Claude write itself? Can it finish two months' work in two hours? Is human presence at the workplace redundant?

[Introduction] The sudden emergence of Claude Cowork is not only a technological miracle of building a system in 10 days but also a cruel interrogation of the value of human occupations: When AI can finish two months' work in two hours, should we rejoice in our liberation or fear being replaced?

The worker version of Claude has made a grand debut, bringing a huge shock to the whole network!

Anthropic's AI office tool - Cowork - has exploded across the network overnight, pushing white-collar workers directly to the edge of unemployment.

With it, one person can boost the efficiency of an entire company.

Some people are saying that Claude Cowork has been seriously underestimated by the whole network!

It can make its own plans, conduct active reasoning, and synchronize progress in real-time. It can instantly organize messy files clearly;

Even scattered notes can be turned into a logically clear report in its hands. It is truly the ultimate productivity booster.

Some big V bloggers have evaluated it as the first step towards a real large-model OS system.

"To be honest, it is AGI! The tax filing work that originally took 40 hours has been significantly compressed to 15 minutes."

Netizens have bluntly said: Claude Cowork is the new gold standard for AI productivity!

Even more astonishingly, the father of Claude Code revealed that all the code for Cowork was written by Claude Code itself.

We have finally entered an era where AI commands and creates itself.

Delivered in 10 days, Claude wrote 100% of the code

This is the most heart-wrenching part: Claude wrote Claude Cowork and completed it in just one and a half weeks (10 days).

At this moment, AI has truly achieved an end-to-end closed loop.

Dario Amodei once said that in the next 3 - 6 months, AI will write 90% of the code. Now, the credibility of this statement is still increasing.

If Claude wrote Cowork, then what can humans still do?

For this, Anthropic engineer Felix Rieseberg explained the main processes the team participated in, which are just three things:

• Make decisions on the general direction, overall architecture, and core product decisions;
• Set rules, define boundaries, and break down tasks for Claude;
• Finally, conduct reviews, approve, and merge.

During the actual "coding" time, each developer has to manage 3 - 8 Claude instances on their screen simultaneously.

They are assigned different roles: Some write front-end interactions, some are responsible for back-end logic, some research technical solutions, and some fix bugs reported on Slack...

All tasks are directly assigned to Claude to complete with one click. Humans just give instructions. In Rieseberg's exact words -

Most of our time now is spent on making decisions by commanding this "Claude army" instead of typing code line by line manually as before.

So, what was the opportunity for the release of Claude Cowork?

The father of Claude's self-narration: The whole company was taken over

Boris Cherny, the father of Claude Code, recalled the end of 2024 when it was still the era of Sonnet 3.5, and AI was far from being as capable of planning and iterating as it is today.

At that time, the first version of Claude Code (Claude CLI) was sent to the internal team for trial use.

A few days later, Cherny walked into the office and saw an astonishing scene -

On colleague Robert's computer terminal, Claude CLI was running, and the screen showed a red - green diff comparison view.

Cherny thought it was not very useful at that time but didn't expect that Robert directly used it for writing and modifying code and even let Claude handle his git operations.

Since then, Anthropic engineers have been using Claude for coding every day, and even data scientists have started using it.

In the next few months, similar scenes played out again and again, like dominoes:

• Designers started using Claude Code to create prototypes, revise copywriting, and solve content problems;
• Finance colleagues used it to build models and make financial forecasts;
• The sales team used it to analyze data from Salesforce and BigQuery;
• User researchers used it to quickly process questionnaire results.

Eventually, the father of Claude Code was no longer surprised. He realized that truly powerful AI tools are never just for writing code.

As Felix Rieseberg said, Claude Code is no longer limited to developers. People who don't understand technology are using it to develop products, and those who understand technology are using it for miscellaneous tasks.

The boundary between the two is becoming blurred at a visible speed.

Actually, it's not just users who have this feeling. Several internal teams at Anthropic have been quietly researching one thing in the past few months -

To turn Claude from a chat - buddy into a real doer.

Boris Cherny suddenly proposed to release a streamlined version of what they had been using internally every day. So, they quickly assembled a small team and set an extremely aggressive deadline: Monday.

This is also the reason for the birth of Claude Cowork - a non - programmer version of Claude Code that lowers the threshold for people to use.

Netizens' actual tests: No way, am I really unemployed?

On the second day after Claude Cowork was launched, netizens were completely plunged into existential panic!

For the first time, this question is so close to us: Are we really going to be unemployed?

A marketer named Vibhu installed Claude Cowork out of curiosity, and then he was completely shocked!

Just two hours later, it completed the following tasks -

• Cleaned up 14 job descriptions that had been in the to - do list since November;
• Developed a Q1 marketing strategy with budget allocation;
• Replied to 47 partner emails that had been ignored;
• Finalized 3 announcement copies that had not been scheduled;
• Even completed the brand tone guide promised to the team six months ago;
• Replied to 23 unread LinkedIn private messages.

This was Vibhu's two - month workload, and now Cowork completed it in two hours!

After seeing this, Vibhu panicked and closed his laptop fearfully.

He opened Slack, trying to pretend to be busy, but he found that he had nothing to do.

He could only take a walk, have a coffee, tidy up his desk, and then take another walk.

Now, he really doesn't know what his job is supposed to be anymore. His schedule is empty. His to - do list is empty. His inbox is also empty.

A developer expert found that just five minutes after Cowork was opened, it directly opened a folder and started working on its own!

It invoked his customized "lease review" skill, automatically generated an executable task list, and then organized the results into a document report on its own.

Witnessing this smooth process, he was simply stunned.

A market observer said, "If you have a virtual assistant, you can fire it now!"

He said, "Claude Cowork is incredible. It reorganized all my downloaded files, prepared notes for my meetings, and sorted and archived all my receipts."

In the following ten - minute video, you can feel how amazing Claude Cowork's performance is.

If you are a day trader who opens and closes positions on the same day, you must understand the pain: Reconcile thousands of transactions every year to ensure all data is accurate and meets the requirements of the IRS.

In the past, you had to spend $700 a year to hire someone to handle these things, but now, just give Cowork a prompt, and it will handle it for you!

A big V blogger said that Cowork helped her successfully delete 1293 screenshots on her laptop desktop. Seeing such a clean desktop, she said, "I'm about to cry. I love you, Claude Cowork!"