Building Claude Code with Boris Cherny

The Genesis of AI in Coding: From Early Experiments to Claude Code [00:00]

• Boris Cherny's early involvement with TypeScript is highlighted, even finding his book translated in Japan.
• He notes the significant advancement of AI, where Claude code generates about 80% of code at Anthropic on average.
• Cherny describes shipping 20-30 pull requests daily, with Claude code writing 100% of them without manual editing.
• The feeling of being "behind as a programmer" due to AI's rapid progress is acknowledged.

"The model is improving so quickly that the ideas that worked with the old model might not work with the new model."

A Transformative Era: The Printing Press Analogy for AI [00:33]

• Cherny compares the current AI moment to the printing press in the 1400s.
• He draws a parallel to the scribes who were displaced but evolved into writers and authors as the market for literature expanded.
• This metaphor suggests a similar shift in the software engineering landscape due to AI.

"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."

Early Life and the Spark of Programming [02:12]

• Cherny's initial introduction to coding involved HTML for selling Pokemon cards on eBay, driven by a desire to increase sales.
• He also programmed TI-83 graphing calculators in middle school to solve math problems, eventually using assembly language to optimize programs.
• This early practical application of coding as a means to build useful things shaped his perspective.

"Coding is a means to build things and to make useful things."

The Startup Journey: Pivots and Product-Market Fit [04:55]

• Cherny emphasizes that startups are rarely linear and often require multiple pivots to find product-market fit.
• He recounts an experience with Agile Diagnosis, a medical software company, where shadowing doctors revealed the product's impracticality due to their time constraints.
• The focus was on outcomes and user needs rather than just technology.

"You always have to figure out what the market wants and what users want. And it's never the thing that you think."

Engineering at Meta: From Groups to Code Quality Leadership [11:36]

• Cherny began at Meta on Facebook Groups, driven by the mission of connecting people.
• He progressed to lead code quality for all of Meta, overseeing platforms like Instagram, Facebook, Messenger, and WhatsApp.
• A key initiative was "Better Engineering," mandated by Mark Zuckerberg, dedicating 20% of engineer time to fixing technical debt.

"By the time I left, I was leading code quality for all of Meta."

The Anthropic "Wow Moment" and the Birth of Claude Code [20:32]

• Upon joining Anthropic, Cherny's first pull request was rejected because he wrote it by hand, being directed to use an AI tool called Clyde, the predecessor to Claude code.
• This experience was a significant "aha!" moment, realizing the AI's capability to generate a working pull request.
• Claude code was initially a side project that evolved rapidly, transforming workflows.

"And this was like September 2024, August, something like that. And I think for me, this was my first fuel hi moment at Anthropic."

Claude Code's Rapid Ascent and Internal Debate [23:08]

• Claude code evolved from a simple terminal chatbot to a sophisticated tool leveraging AI's ability to use tools.
• An internal debate occurred at Anthropic about whether to release Claude code due to its productivity-enhancing capabilities.
• The decision to release was driven by the need to study AI safety in the wild.

"Product exists so that we can serve research better and so we can make the model safer."

The Evolving Developer Workflow: 80% AI, 0% Manual Edits [31:00]

• Cherny states that Claude code now writes about 80% of the code at Anthropic on average, and 100% of his code.
• The advent of Opus 4.5 marked an instant shift, leading him to uninstall his IDE as it was no longer needed.
• He reports shipping 20-30 pull requests daily with 100% AI-generated code and zero manual edits, with the AI introducing fewer bugs than he would have.

"Opus 4.5 and Quad Code wrote 100% of every single one. I didn't edit a single line manually."

Rethinking Code Review in the Age of AI [40:47]

• Cherny previously automated tedious work by writing lint rules based on common issues identified in code reviews.
• Now, Claude code runs tests locally and performs initial code reviews, catching about 80% of bugs.
• Human review remains essential for the final pass and approval, ensuring safety and quality.

"Every pull request at Anthropic is code reviewed by Quad Code."

The Generalist Future: Adaptability and Cross-Disciplinary Skills [56:03]

• Anthropic's "member of technical staff" title reflects a generalist approach, encouraging engineers to engage in product, design, and research.
• This trend is accelerated by AI, blurring traditional role boundaries and rewarding multi-disciplinary skills.
• The ability to context-switch rapidly and manage multiple AI agents is becoming increasingly valuable.

"I think the next, you know, billion dollar product, you know, like after Quad Code, whatever the next startup is that, you know, becomes the next trillion dollar startup, it might just be like one person that has some cool idea and their brain just is able to think across, you know, engineering and product and business."

A New Era of Software Creation: Prototyping over Documentation [01:00:28]

• The culture at Anthropic emphasizes "showing" rather than "writing" (e.g., PRDs), with rapid prototyping being key.
• Features like agent teams were developed through hundreds of prototypes to refine the user experience.
• This iterative, hands-on approach is enabled by the low cost of building and the uncertainty of outcomes, requiring an exploratory mindset.

"On our team the culture is we don't really write stuff. We just we show."

The Printing Press Analogy Revisited: The Evolution of Scribes [01:25:48]

• Cherny reiterates the printing press analogy, highlighting how AI is making sophisticated tasks accessible to a wider audience.
• He compares the current situation to medieval scribes being replaced by authors as the printing press expanded the market for written content.
• This suggests that AI will not eliminate software engineers but will transform their roles and create new opportunities.

"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."

Evolving Skills: Adaptability in a Rapidly Changing Landscape [01:31:56]

• Cherny reflects on the need for intellectual humility and adaptability as AI models evolve, making previously "bad" ideas viable.
• He notes that while deep coding skills might become less critical, methodical thinking, curiosity, and cross-disciplinary capabilities are more valuable.
• The future rewards generalists who can navigate diverse domains and adapt quickly to new technologies and approaches.

"The stuff that's left behind is uh best left behind is maybe like very strong opinions about like code style and languages and things like this."

The Future of Software Engineering: A Call for Adaptability and Generalism [01:35:13]

• The conversation concludes with the understanding that constant adaptation is key, as new AI models will continue to reshape the field.
• Cherny recommends books like "The Three-Body Problem" series for science fiction enthusiasts, "Accelerando" for its depiction of rapid technological advancement, and "Functional Programming in Scala" for its insights into better coding practices.

"The one certain thing we can be sure is whenever the next model comes out, it'll change again. And you need to be curious and open to adapting how you work, right?"