← All talks Head of Claude Code: What happens after coding is solved | Boris Cherny 1:27:45

Lenny's Podcast

Head of Claude Code: What happens after coding is solved | Boris Cherny

Anthropic's Claude Code team reports 200% higher pull requests per engineer and 100% of Boris Cherny's code now written by AI since November.

BC

Boris Cherny

Head of Claude Code · Anthropic

LR

Lenny Rachitsky

Host · Lenny's Podcast

The numbers, across the talk20 moments · 1:27:45

4% · 00:05:27 2x DAU in one month · 00:01:08 100% · 00:16:12 ~20% (Feb) · 00:15:04 +200% · 00:20:29 a few % per year · 00:21:06 hundreds of thousands $/mo · 00:27:44 <1% · 00:32:59 ~70% · 00:33:41 70% · 00:44:17 55% · 00:45:01 40% · 00:47:30 60% · 00:48:13 10 days · 00:53:06 15–30 seconds · 01:07:36 ~80% · 01:09:30 $2B · 01:15:10 48 of 50 · 01:23:20 ~50% overlap · 00:42:17 10–20% · 01:05:02

0:001:27:45

01 · AI-authored code share on GitHub

GitHub / Claude Code

A semi-analysis report found Claude Code now authors 4% of all public GitHub commits, with a prediction of a fifth of all commits by end of year. Boris notes private repos are likely higher.

Share of GitHub commits

“There's this report that recently came out that I'm sure you saw by semi-analysis that showed that 4% of all GitHub commits are authored by Claude Code now.”

▲20% (a fifth)

Predicted commit share by EOYprojected

“And they predicted it'll be a fifth of all code commits on GitHub by the end of the year.”

02 · Claude Code growth

Anthropic

Claude Code's growth is still accelerating — daily active users doubled in the past month — and Anthropic raised at a valuation over $350 billion.

▲2x DAU in one month

DAU growth in a month

“Just in the past month, their daily active users has doubled.”

Anthropic valuation

“They just raised around it over $350 billion.”

03 · Personal coding fully delegated to AI

Anthropic

Boris says 100% of his code is written by Claude Code, he hasn't hand-edited a line since November, and he ships 10–30 pull requests a day. Claude also reviews 100% of pull requests at Anthropic.

Code written by AI

“So 100% of my code is written by cloud code.”

▲10–30 PRs/day

PRs shipped per dayestimated

“And so every, you know, every day I ship like 10, 20, 30, 30 requests, something like that.”

PRs reviewed by Claude

“So here at Anthropic Cloud reviews 100% of whole requests.”

Share of code AI-written in Febestimated

“Because even in February when we released it, it was writing maybe, I don't know, like 20% of my code, not more.”

Share of code AI-written in Mayestimated

“And even in May, it was writing maybe 30%.”

04 · Engineering productivity gains

Anthropic

Since introducing Claude Code, Anthropic roughly 4x'd its engineering team while pull requests per engineer rose 200% — versus the few percentage points per year Boris saw with hundreds of engineers working on code quality at Meta.

PRs per engineer

“But product. per engineer has increased 200% in terms of like full requests.”

Engineering team growthestimated

“We probably like 4x the engineering team or something like this.”

▲a few % per year

Prior-era annual gain (Meta)estimated

“With hundreds of engineers working on it, you would see a gain of like a few percentage points of productivity, something like this.”

05 · Token spend per engineer

Anthropic

Boris says some Anthropic engineers now spend hundreds of thousands of dollars a month on tokens, and advises companies to be generous with tokens before optimizing.

▲hundreds of thousands $/mo

Token spend per engineerestimated

“You know, at Anthropic, we're starting to see some engineers that are spending, you know, like hundreds of thousands a month in tokens.”

06 · Historical analog: the printing press

Boris uses the printing press as an analog: literacy under 1% in mid-1400s Europe, printing costs falling ~100x over 50 years, and literacy reaching ~70% globally over 200 years.

Literacy in mid-1400s Europe

“literacy was actually very low. There was sub 1% of the population.”

▼100x cheaper

Printing cost declineestimated

“It went down something like 100x over the next 50 years.”

Literacy 200 years laterestimated

“But over the next 200 years, it went up to like 70% globally.”

07 · Job satisfaction after adopting AI tools

Lenny's informal Twitter polls found 70% of engineers and PMs enjoy their job more since adopting AI tools, versus 55% of designers, with 10% and 20% enjoying it less respectively.

Engineers/PMs enjoying job more

“And both engineers and PMs, 70% of people said they are enjoying their job more.”

Engineers/PMs enjoying job lessestimated

“And about 10% said they're enjoying their job less.”

Designers enjoying job more

“Designers, interestingly, only 55% said they're enjoying their job more and 20% said they're enjoying their job less.”

08 · Latent demand product signals

Facebook / Meta

Boris cites Facebook Marketplace and Dating as latent-demand examples: 40% of Facebook group posts were buying and selling, and 60% of profile views were between non-friends of the opposite gender.

Group posts buying/selling

“that 40% of posts in Facebook groups are buying and selling stuff.”

Cross-gender non-friend profile views

“60% of profile views are people that are not friends with each other that are opposite gender.”

09 · Building Cowork with Claude Code

Anthropic

Anthropic's team built the Cowork product — including a full virtual machine and security guardrails — entirely with Claude Code in about 10 days.

Build time for Cowork

“It was fully implemented with Quad Code. It took about 10 days.”

10 · Autonomous agent run length

Anthropic

Model autonomy has grown from 15–30 seconds of unattended running with Sonnet 3.5 a year ago to 10–30 minutes on average with Opus 4.6, and in some cases hours, days or weeks.

●15–30 seconds

Unattended run time (a year ago)estimated

“it could run for maybe 15 or 30 seconds before before it started going off the rails.”

▲10–30 minutes

Unattended run time (Opus 4.6)

“But nowadays with Opus 4.6, you know, on average, it'll run maybe 10, 30, 20, 30 minutes unattended.”

11 · Plan mode usage

Anthropic

Boris starts about 80% of his Claude Code tasks in plan mode, which is implemented by injecting a single sentence into the model's prompt.

Tasks started in plan modeestimated

“I start almost all of my tasks in plan mode, maybe like 80%.”

12 · Anthropic revenue scale

Anthropic

Lenny cites publicly stated figures of roughly $2B in revenue for Claude Code and $15B for Anthropic overall, while Boris notes most of the world still doesn't use AI.

Claude Code revenueestimated

“I think Claude alone is making $2 billion in revenue.”

Anthropic revenue

“You think Anthropic, I think the number you guys put out, you're making $15 billion in revenue.”

13 · Evaluating Cowork against non-technical use cases

Anthropic

Anthropic used Lenny's list of 50 non-technical Claude Code use cases as an eval; Cowork was released once it could handle 48 of the 50.

“And I think at the point where we hit where coworker was able to do like 48 out of the 50. They were like, okay, it's pretty good.”

14 · Role convergence across product teams

Anthropic

Boris observes roughly 50% overlap between engineering, design and product management roles on the Claude Code team, where everyone codes.

●~50% overlap

Role overlapestimated

“One thing that we're starting to see is there's maybe a 50% overlap in these roles where a lot of people are actually just doing the same thing.”

15 · Scaffolding vs. general models

Boris says added scaffolding around a model typically improves performance only 10–20%, and those gains are usually wiped out by the next model release.

Gain from scaffoldingestimated

“maybe scaffolding can improve performance, maybe 10, 20%, something like this. But often these gains just get wiped out with the next model.”