New Intelligence Report
Anthropic's Claude Cowork has just officially launched! Moreover, Boris Cherny, the creator of Claude Code, openly shared his personal configuration in a 40-minute interview, delivering a series of hardcore insights that left the online community thrilled!
Anthropic's Claude Cowork has set the entire AI world ablaze.
Today, Anthropic is valued at $1 billion.
And today, another major news broke: Claude Team and Enterprise versions are now live!
Although Cowork is still in the research preview stage, the official release includes two critical capability upgrades.
1. @Mention Projects
In Cowork, you can directly inject complete background information into a meeting or collaborative task by @mentioning a project.
From now on, we no longer need to repeatedly explain the project's background; AI can understand our goals, constraints, and existing materials.
Every discussion advances on the basis of "known context," just as naturally as developers @mentioning a module in a code repository.
2. Support for Real-time Screen Capture During Runtime
Now, when using Claude in Chrome, Cowork can display real-time screen captures during operation.
This makes AI no longer an assistant that "can't see what you're doing," but one that can understand your current page state, follow your actual operation pace, and provide instant suggestions at critical steps.
This step transforms AI from a "consultant in a chat box" into a true work partner.
With Cowork, we can introduce new suppliers on a large scale.
In the supplier introduction process, Cowork can help us automatically organize candidate lists, compare supplier capabilities, quotes, and risks, and generate evaluation tables and decision recommendations.
This compresses a process that originally required multiple meetings and Excel into a collaborative AI workflow.
Additionally, we can use Cowork for industry market size assessments.
In this process, Cowork can aggregate public industry information, structurally analyze market size and growth potential, and output conclusions and scenarios directly usable for reporting.
From "researching information" to "forming judgments," there is no longer a need to repeatedly switch tools.
The Art of Commanding a One-Person Army
Claude Code's Creator Reveals His Personal Configuration
Another major news today is that Boris Cherny, the creator of Claude Code, appeared on the interview program of renowned tech podcast host Greg Isenberg.
This interview, hailed as a "master-level private coaching session," went viral immediately after its release!
In 42 minutes, Boris unreservedly shared hardcore insights about Claude Cowork and Claude Code, leaving the audience thrilled.
As long as you know how to use it, you can outperform 99% of the people on this planet.
This is Greg Isenberg's opening evaluation of Claude Cowork.
But in this interview, what was even more "mind-blowing" than this statement was the fact Boris revealed:
As the creator of the tool, in the past two months, 100% of the code he submitted was written by AI; he didn't write a single line of code by hand.
From basic tasks like organizing receipts, making tables, and controlling browsers, to completely deconstructing his own viral "divine workflow" across the internet, Boris did not stay at the theoretical level but provided a practical guide that is truly ready to be implemented.
Next, let's see what "black technology" Boris actually demonstrated and what the popular "configuration" looks like.
TL;DR Version
Core Intelligence: What Exactly Did Boris Cherny Reveal?
In this "$1 billion" configuration, here are the three most valuable and most "copyable" technical points in Boris's workflow.
When AI Takes Over the Desktop, It's More Magical Than Magic
Without further ado, Boris shared his screen and demonstrated Cowork's practical capabilities.
The demonstration scenario was very down-to-earth: processing messy invoices.
First Challenge: Taming Chaos, It Even Learned to "Ask Back"
There is a folder named "Receipts" on Boris's desktop, containing several messy receipt images.
He opened the folder's permissions to Cowork and gave the instruction: "Rename these files so that the filenames match the dates on the receipts."
At this point, Cowork demonstrated an interesting trait—Reverse Elicitation.
One receipt had a blurry date, and Cowork stopped and asked Boris: "This one is unclear. Do you want me to skip it, or will you decide?"
Once confirmed, all files were neatly renamed in an instant.
Second Challenge: A Spectacular "Cloud Migration" Operation
The next operation was even more jaw-dropping.
Boris said: "Make a table out of these receipts."
Cowork generated a CSV file locally in a few seconds.
But Boris deliberately challenged it: "I don't want it locally; get me a Google Sheet."
At this moment, a magical scene occurred.
Cowork actually opened the Chrome browser (of course, this requires user authorization), logged into Google Sheets, created a new spreadsheet, and then filled in the data one by one.
Although there was a minor formatting issue during the process—the data segmentation was not done correctly.
But before Boris could say anything, Cowork realized it and immediately started correcting it.
Third Challenge: I Drink Coffee, Let 10 "Clones" Do the Work for Me
The table was done, and Boris added another sentence: "Help me open Gmail and send this table to Amy."
Cowork skillfully opened the email, found Amy from the contacts, wrote a draft, and waited for Boris to click send.
However, sending emails is just basic operation. Boris also gave a more extreme workplace case—Slack Automated Follow-ups.
He had Cowork regularly monitor the team's project progress table.
Whenever it found a column of data not filled in, Cowork would automatically go to Slack and "privately message" the corresponding engineer to follow up.
Previously, this required the project manager to chase each person one by one; now, AI handles it automatically.
In Boris's view, current AI is like a junior employee, sometimes a bit slow (like clicking on web pages), but the real killer feature is—Parallelism.
For example, in the dozens of seconds Cowork was helping him send emails, Boris had already switched to another tab to have another Cowork research "which startup podcasts are worth listening to."
"I won't foolishly wait for it to work," Boris said. "I will open 5 to 10 such task windows simultaneously."
My daily work now is not "writing code," but "taking care" of this group of Claudes.
They are working, while I drink coffee or deal with other logic.
Boris Personally Teaches the Divine Workflow
Boris previously shared his Claude Code configuration on Twitter, which was retweeted and bookmarked nearly 100,000 times.
In the program, he detailed this solution known as "efficiency surpassing 99% of humans."
Full version: https://x.com/bcherny/status/2007179832300581177
1. Shadow Clone Technique: Don't Wait Foolishly, Parallelism is the Way
Don't stare at AI thinking.
Today's engineers are no longer typists but "multithreaded task managers."
As mentioned earlier, Boris's habit is to run 5 to 10 Claude tasks simultaneously. And work simultaneously on the terminal, web, and even mobile app.
When one task is in the planning phase, he switches to the next task.
This "multithreaded workflow" can make your output grow exponentially.
2. Refuse "Intelligence Reduction": Use Only the Smartest Brain
Many people use small models to save money, which is actually a misunderstanding.
Boris strongly recommends always using Opus 4.5 with Thinking mode enabled.
Although a single call is expensive, because it is smart enough to get things right in one go, with minimal rework, it ultimately saves more tokens and is more cost-effective.
3. Build a Team's "Second Brain": Let AI Stop Making the Same Mistakes
This is an extremely simple text file, but its role is huge.
Boris's team places a claude.md file in the project root directory.
Claude made a mistake? Don't just scold it; write "don't do this next time" into this file. Have special code specifications? Write them in too.
This is Claude's memory library. Every time it works, it will read this file first to ensure it doesn't make the same mistake again.
4. Grand Summoning Technique: @Claude in Comments to "Forge" Code Directly
This trick is used by Boris's team with great satisfaction.
If they find a problem during code review or want to update claude.md, they will directly @Claude in the comments.
Then, AI will automatically do the work, even pushing it directly to the branch.
This is called "compound engineering," allowing the system to evolve automatically over time.
5. Plan Before Acting: Planning Mode is the True Divine Skill
Most of the time, Boris enters "Planning Mode" first.
For example, "I want to write a PR, what do you think?"
Then, he will discuss back and forth with Claude until the plan is perfect.
Once the plan is determined, switch to execution mode, paired with the Opus 4.5 model, the code can usually be completed in one step and run directly.
Remember this sentence:
As long as the plan is good, the code is good.
6. Give It "Eyes": Let Claude Verify Itself
This is the key trick to make the effect take off: Give Claude a way to verify its own work results.
Just as a painter cannot paint with their eyes covered, an AI writing code needs to see the results.
Install a Chrome extension for it so it can run code and see the pages in the browser.
If it can self-test, the quality of its output will be astonishingly good.
The Era of Hand-Writing Code is Over
When the host asked about the outlook for the next year, Boris's answer was thought-provoking.
Last year, I and the company's founder Dario predicted that by the end of this year, no one would be writing code anymore.
At the time, no one believed it, thinking it went against intuition.
But if you believe in exponential growth, this is inevitable.
Now, I have achieved 100% AI-generated code.
As Boris said, "If you had asked me a year ago, I would never have dreamed that I would be writing code in this way."
For Cowork, this is just the beginning.
The future way of work will no longer be you learning complex tools, but you hiring AI Agents, each with specific Skills, to handle all the trivial matters from AutoCAD drawing to invoice reimbursement.
The era of manually writing code is over. The new game rule is:
Whoever can better "command" this AI army is the winner.
One More Thing
Additionally, there is another good news today: Claude has officially entered Excel Pro!
Seriously, Claude's entry into Excel Pro this time finally feels like it "understands the working class."
Previously, using AI in Excel always felt like a trial version: files had to be uploaded one by one, and generated content might accidentally overwrite your hard-made tables, making it nerve-wracking to use.
This update, Claude directly filled these pain points!
Files can be dragged directly, multiple files can be dropped together; when writing results, it automatically avoids existing cells, no longer acting as a "table assassin"; even long conversations that last for half an hour will have their context automatically compressed.
These may seem like "small changes," but each one hits the nerves of heavy Excel users.
In one sentence: this is a truly user-centric upgrade for the table-loving crowd.
References:
https://x.com/claudeai/status/2014748925144793476
https://www.youtube.com/watch?v=DW4a1Cm8nG4
Follow ASI Instantly
⭐Like, Share, and View - One-Click Triple Action⭐
Light up the star mark to lock in New Intelligence's fastest push!