Claude now writes its own codebase

**SPEAKER_1** (0:00)
Study and play. Come together on a Windows 11 PC. And for a limited time, college students get the best of both worlds. Get the Unreal College Deal, everything you need to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 Premium, and a year of Xbox Game Pass Ultimate with a custom color Xbox Wireless Controller. Learn more at windows.com/studentoffer. While Supplies Last ends June 30th. Terms at aka.ms slash college PC.

**SPEAKER_2** (0:31)
When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen, and helps reach people with the right skills, certifications, and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsored job credit at indeed.com/podcast.
Terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs.

**SPEAKER_3** (1:00)
This Father's Day, do more with dad and spend less with low prices guaranteed at the Home Depot. Get him fired up with a new grill and accessories, like the next grill five burner for just $299, so you can spend more time together while he becomes the grill master he was always meant to be. Or build memories with savings on top brand power tools, so you can tackle projects side by side. Gift more and do more together this Father's Day with help from the Home Depot. Exclusions apply to homedepot.com/pricematch for details.

**Stage Zero** (1:30)
Anthropics Claude model is now writing over 80 percent of the company's own code base.

**SPEAKER_5** (1:36)
Which is, I mean, that is a volume of automated work that is really difficult to contextualize when you just hear the number.

**Stage Zero** (1:42)
Yeah.

**SPEAKER_5** (1:43)
You really have to place it right next to their other public moves to see the full picture. Because they published those internal development metrics at the exact same moment they proposed a global coordinated pause on artificial intelligence development.

**Stage Zero** (1:56)
Right. The timing is wild.

**SPEAKER_5** (1:58)
Exactly. And all of that is happening while they simultaneously file for an initial public offering, which is tied to evaluation nearing $1 trillion.

**Stage Zero** (2:08)
So that sequence of events, it forces a very specific question for anyone who's actually watching this industry.
If the machine is already heavily involved in building itself, is a coordinated global pause actually possible to execute? Yeah. Or is proposing a pause serving an entirely different strategic purpose for the company?

**SPEAKER_5** (2:26)
Well, to figure that out, you really have to look closely at what that automation actually looks like on the engineering floor. Because the abstract concept of a machine writing code is one thing, but the actual mechanical changes happening inside their development teams, that is what matters here.

**Stage Zero** (2:42)
So, if we look at those internal metrics, they show a shift in the fundamental process of software creation. Engineers working there are, they're currently merging eight times as much code on a daily basis compared to their historical long-term baseline.

**SPEAKER_5** (2:58)
Eight times.

**Stage Zero** (2:59)
Eight times, yeah.
But the volume is really only part of it. They also measure the duration of tasks a model can complete autonomously, and that duration is doubling over highly compressed intervals.

**SPEAKER_5** (3:10)
Which is the crazy part.

**Stage Zero** (3:12)
We're talking about a shift from a model handling tasks that require a few minutes of effort to managing tasks that demand a half day of continuous focused human effort.

**SPEAKER_5** (3:22)
Yeah, but writing code and reviewing code are two entirely different disciplines.

**Stage Zero** (3:26)
Right.

**SPEAKER_5** (3:26)
Producing eight times the volume of code looks amazing on a spreadsheet, but you have to filter that through Amdahl's law. In any complex system, your overall speed up is strictly limited by the portion of process that cannot be parallelized or automated. Okay. So if you accelerate the generation phase to some extreme degree, you do not necessarily build the product faster, you just move the bottleneck.

**Stage Zero** (3:50)
Because someone still has to read it.

**SPEAKER_5** (3:51)
Exactly. The constraint shifts from the person writing the code to the human reviewers who have to read, understand and actually approve this massive new tidal wave of generated text. And I mean, if you are an engineer, you know that reading someone else's code is usually way harder than writing your own.

**Stage Zero** (4:10)
Oh, for sure.

**SPEAKER_5** (4:10)
Now imagine reading code written by a machine that generates it a hundred times faster than you can even read it.