Boris Cherny, Creator of Claude Code

Timestamps

Key Arguments

The Printing Press Analogy (85:00)

Boris compares the current AI moment to the invention of the printing press. Before the press, less than 1% of Europe was literate. Scribes were a tiny elite employed by kings who were often illiterate themselves. The press dropped the cost of printed material 100x in 30 years and increased quantity 10,000x in a century. The scribes didn't disappear — they became writers and authors, and the entire market for written work expanded beyond prediction. Software engineers are today's scribes; coding is becoming accessible to everyone, and the market for what can be built will expand enormously.

Why Claude Code Uses Agentic Search Over RAG (55:00)

The team tried local vector databases, recursive model indexing, and various retrieval approaches. RAG had problems: code drifted out of sync with the index, and permissioning the index raised serious privacy/security concerns. In the end, agentic search (glob + grep) outperformed everything. Boris credits his Instagram experience where click-to-definition was broken, so engineers searched with patterns like foo( — the same approach works well for models.

Safety as Swiss Cheese Model (52:00)

There's no single perfect answer for agentic safety. Anthropic uses layered defenses: model alignment (Opus 4.6 is most aligned), runtime classifiers that block prompt-injected requests, and sub-agent summarization (e.g., web fetch results are summarized by a sub-agent before returning to the main agent). For code, every PR gets Claude code review first (~80% of bugs caught), then human review.

The Year of the Generalist (90:00)

At Anthropic, everyone codes — engineers, designers, data scientists, finance. The flat "Member of Technical Staff" title encodes this expectation. Boris predicts the next trillion-dollar startup may come from one person who thinks across engineering, product, business, and design. Skills to leave behind: strong opinions about code style, languages, and frameworks. Skills that matter more: being methodical and hypothesis-driven, curiosity, adaptability, and context-switching ability.

How Claude Code Actually Ships (42:00)

Boris writes 10-20 PRs daily. Opus 4.5 with thinking mode wrote 100% of every single one — he didn't edit a single line manually. The team does dozens to hundreds of prototypes before shipping a feature. No PRDs — "better send a PR." Cat Woo, their product lead, is extremely technical and prefers working prototypes over documents.

Uncorrelated Context Windows and Agent Teams (75:00)

Sub-agents have fresh context windows that don't know about the parent context. This "uncorrelated" approach, combined with throwing more tokens at the problem, is a form of test-time compute that produces better results. Agent teams clicked with Opus 4.6, where agents have "cute exchanges" discussing problems. Best used for complex tasks where a single Claude struggles.

Predictions Made

• Everyone will code — Non-engineers (designers, data scientists, finance) will routinely write code as AI makes it accessible (Confidence: high)
• The next billion-dollar product may be built by one person — A generalist who thinks across disciplines (Confidence: medium)
• Deep coding skills may not be required in 6 months — Debugging methodology still matters now but the window is closing (Confidence: medium)
• Coding language/framework debates will become irrelevant — Models can rewrite between languages trivially (Confidence: high)

Notable Quotes

"Now we're at the point where Claude Code writes, I think something like 80% of the code at Anthropic on average. I wrote maybe 10-20 PRs every day. Opus 4.5 and Claude Code wrote 100% of every single one. I didn't edit a single line manually." — Boris Cherny at 00:00

"The model just wants to use tools. If you give it a tool, it will figure out how to use it to get the thing done." — Boris Cherny at 35:00

"Don't try to put the model in a box. Don't try to force it to behave a particular way. Let it do its thing." — Boris Cherny at 36:00

"The scribes didn't disappear. They became writers and authors. And the reason they exist is because the market for literature just expanded a ton." — Boris Cherny at 87:00

"I think the next trillion-dollar startup might just be one person that has some cool idea and their brain is able to think across engineering and product and business." — Boris Cherny at 92:00

Resources Mentioned

• Accelerando by Charles Stross — "essentially the product roadmap for the next 50 years"
• Functional Programming in Scala — "teaches you how to think in types"
• Liu Cixin's short stories (Three-Body Problem author)
• Claude Cowork — Anthropic's new product bringing agentic AI to non-engineers

Connections

• boris-cherny-on-what-grew-his-career-and-building-at-anthropic - Earlier interview covering Boris's Meta career and TypeScript book in more depth
• im-boris-and-i-created-claude-code - Boris's own tips thread on Claude Code productivity (parallel worktrees, CLAUDE.md philosophy)
• thread-based-engineering-how-to-ship-like-boris-cherny - Deep dive into Boris's shipping methodology
• how-ai-is-transforming-work-at-anthropic - Broader look at Anthropic's engineering culture
• ai-codes-better-than-me-now-what - The identity crisis Boris describes around coding skills being commoditized
• claude-code-best-practices - Practical patterns that align with Boris's described workflow