[Host]: You were the first ever TypeScript book with O'Reilly.

Boris Cherny: Yeah, I found that book translated in Japanese in this little town in Japan. That was just the coolest moment. And then I realized I don't remember TypeScript at all. Now we're at the point where Quad Code writes, I think something like 80% of the code had Enthropic on average. I wrote maybe 10 20 p requests every day. Opus 4.5 and Quad Code wrote 100% of every single one. I didn't edit a single line manually.

[Host]: Andre Karpathy posted that he's never felt as much behind as a programmer as he is now.

Boris Cherny: This is something I really struggle with. The model is improving so quickly that the ideas that worked with the old model might not work with the new model. One metaphor I have for this moment in time is the printing press in the 1400s because there was a group of scribes that knew how to write.

[Host]: Some of the kings were illiterate who are employing the scribes.

Boris Cherny: And if you think about what happened to the scribes, they ceased to become scribes, but now there's a category of writers and authors. These people now exist. And the reason they exist is because the market for literature just expanded a ton.

Boris Cherny: It starts a while back. So, when I was maybe 13 or something like this, I started selling my old Pokemon cards on eBay. And I realized that on on eBay, you can actually like write HTML... Then the second thing was... we had these old TI83 graphing calculators... I wrote these little programs to just program the answers... then the math got more advanced... so I had to drop down from basic to assembly to just make the program run a little bit faster.

[Host]: Oh wow. So like in high school you dropped down to assembly.

Boris Cherny: Then the thing I realized is everyone in my class was starting to realize that I had the solver... so I bought this little serial cable... then the next math test, everyone on the class just got A's... But for me, it was very practical. Coding is a means to build things and to to make useful things.

Boris Cherny: Yeah, so I started on Facebook groups... I was really excited about it because of this mission of connecting people to their community... Then I spent a few years at Instagram after... what I very quickly realized is that I was just not effective at working on the stack because it was such a terrible stack and so I just went and started working on Dev Infra... By the time I left, I was leading code quality for all of Meta.

[Host]: Zuck mandated that every engineer at the company 20% of their time has to be spent fixing tech debt.

Boris Cherny: And we called this better engineering... quality actually contributes like you know double digit percent to to productivity. It turns out even even at the biggest scale... As a model, you might just pick the wrong thing and then, you know, like the user has to course correct you.

Boris Cherny: When I joined Anthropic, I did a bunch of ramp up projects... and I wrote my first pull request by hand because I thought that's how you write code... But even at the time... there was this thing called Clyde... Adam rejected my PR and he was like, "Actually, you should use this Clyde thing for it instead." ...it sped out a working PR. It just one-shotted it.

Boris Cherny: I just wanted to, you know, hack something up quite quickly cuz we didn't have quad code back then... and I gave it a single tool which was the bash tool... and I asked it you know like I actually didn't know if it could even do this but I asked it like what music am I listening to and uh it just wrote a little Apple script program... This is one of the corollaries is just let the model do it do its thing. Don't try to put it in a box.

Boris Cherny: So for me, the way that I do it generally is I have five terminal tabs. Each one of them has a checkout of their repository... Almost every time I start in plane mode... Once there is a good plan, it just it will oneshot the implementation almost every time... I just uninstalled my ID cuz cuz I just didn't need it at that point.

[Host]: Shipping this much code... The obvious question that comes up is well the review.

Boris Cherny: So today the way this looks is... every pull request at Enthropic is code reviewed by quad code. Uh and that actually catches maybe like 80% of bugs... There's always an engineer that does the second pass of code review. Um and you know there there always has to be a person in the loop approving the change.

Boris Cherny: We launched agent teams earlier this week. This is our implementation of swarms... You have a bunch of different uncorrelated context windows... It's uncorrelated because the main agent prompts the sub agent but the sub agents context window is fresh... it turned out that agentic search just outperformed everything and when I say agentic search, this is a fancy word for glob and grap. That's all it is.

Boris Cherny: Okay, so the stuff that's left behind is maybe like very strong opinions about like code style and languages... Other skills that I think are more valuable are being curious and being open to doing things beyond your swim lane... So in in some ways I think this will be the year of the generalist.

Boris Cherny: And then on the technical side, I would strongly recommend functional programming in Scola... it'll just teach you how to think in types... For people that are new to sci-fi... I really love Accelerondo by St. This is a book I would totally recommend. It's like essentially the product roadmap for the next 50 years.