Building Claude Code with Boris Cherny

The discussion centers on the evolution and impact of AI coding agents, particularly Claude Code. Boris Cherny, creator of Claude Code and former Meta engineer, recounts his journey from early coding experiences to leading AI-driven code development. He details Claude Code's transition from a side project to a widely adopted developer tool, including internal debates about its public release and the emphasis on safety. Cherny shares his current AI-assisted workflow, highlighting the shift towards parallel processing and AI-driven code review. He draws parallels between the transformative impact of AI on software engineering and the printing press's effect on literacy, emphasizing the need for adaptability and a generalist skillset in the age of AI.

Outlines

Part 1: Origins, Career, and Philosophy

Part 2: The Meta Years: Infrastructure and Quality

Part 3: Anthropic and the Birth of Claude Code

Part 4: AI-Driven Development Workflow

Part 5: Automation, Review, and Safety Layers

Part 6: Anthropic Culture and Internal Processes

Part 7: Expanding the Ecosystem: Cowork and Swarms

Part 8: The Future of Engineering and Identity

Part 9: Final Reflections and Advice

Sign in to continue reading, translating and more.

Open full episode in Podwise

The Pragmatic Engineer

Part 1: Origins, Career, and Philosophy

00:00From Rejected Pull Request to Claude Code: An AI Coding Tool's Origin

From Rejected Pull Request to Claude Code: An AI Coding Tool's Origin

00:58From Pokemon Cards to Weed Reviews: Boris Cherny's Early Coding Ventures

From Pokemon Cards to Weed Reviews: Boris Cherny's Early Coding Ventures

03:33Startup Pivots and Product-Market Fit: Lessons from Agile Diagnosis

Startup Pivots and Product-Market Fit: Lessons from Agile Diagnosis

07:32Outcome-Focused Engineering: Balancing Technical Depth with Practical Application

Outcome-Focused Engineering: Balancing Technical Depth with Practical Application

Part 2: The Meta Years: Infrastructure and Quality

10:17From Facebook Groups to Instagram in Japan: Connecting People and Overcoming Time Zones

From Facebook Groups to Instagram in Japan: Connecting People and Overcoming Time Zones

13:53Instagram's DevInfra Challenges: Migrating from Python and Improving Code Quality

Instagram's DevInfra Challenges: Migrating from Python and Improving Code Quality

16:10Meta's Better Engineering Program: Measuring the Impact of Code Quality

Meta's Better Engineering Program: Measuring the Impact of Code Quality

Part 3: Anthropic and the Birth of Claude Code

18:46Joining Anthropic: A First Pull Request Rejected and a Wow Moment with AI

Joining Anthropic: A First Pull Request Rejected and a Wow Moment with AI

20:48Claude Code's Origin: From Chatbot to Agentic AI for Engineers

Claude Code's Origin: From Chatbot to Agentic AI for Engineers

24:01The Bitter Lesson: Letting the Model Do Its Thing with Tool Use

The Bitter Lesson: Letting the Model Do Its Thing with Tool Use

27:02Safety First: Anthropic's Decision to Release Claude Code for Research

Safety First: Anthropic's Decision to Release Claude Code for Research

28:50Internal Adoption and the Question of Forcing Users: Claude Code's Growth

Internal Adoption and the Question of Forcing Users: Claude Code's Growth

Part 4: AI-Driven Development Workflow

30:44Opus 4.5 and the Instant Switch: Trusting AI to Write All the Code

Opus 4.5 and the Instant Switch: Trusting AI to Write All the Code

32:47Parallelism and Terminal Tabs: Boris Cherny's Development Workflow

Parallelism and Terminal Tabs: Boris Cherny's Development Workflow

35:10Coding on a Phone: Parallel Agents and the Shift in Workflow

Coding on a Phone: Parallel Agents and the Shift in Workflow

36:14Plan Mode and One-Shot Implementations: Maximizing Productivity with Claude

Plan Mode and One-Shot Implementations: Maximizing Productivity with Claude

Part 5: Automation, Review, and Safety Layers

39:27Code Review in the Age of AI: Automating Tedious Work and Ensuring Human Oversight

Code Review in the Age of AI: Automating Tedious Work and Ensuring Human Oversight

42:22Code Review for Every Project? Balancing Automation with Human Oversight

Code Review for Every Project? Balancing Automation with Human Oversight

44:12Writing Lint Rules with Claude: Deterministic Steps and Best of N

Writing Lint Rules with Claude: Deterministic Steps and Best of N

47:04Safety in Layers: Prompt Injection and the Swiss Cheese Model

Safety in Layers: Prompt Injection and the Swiss Cheese Model

48:41RAG vs. Agentic Search: The Evolution of Retrieval in Claude Code

RAG vs. Agentic Search: The Evolution of Retrieval in Claude Code

51:07Permissioning Complexity: Classifiers, Static Analysis, and Allow Lists

Permissioning Complexity: Classifiers, Static Analysis, and Allow Lists

Part 6: Anthropic Culture and Internal Processes

53:34Member of Technical Staff: The Title-less Culture at Anthropic

Member of Technical Staff: The Title-less Culture at Anthropic

55:34The Expanding Roles: Data Scientists and the Power of Claude Code

The Expanding Roles: Data Scientists and the Power of Claude Code

57:59No PRDs, Just Prototypes: Building and Showing Instead of Writing

No PRDs, Just Prototypes: Building and Showing Instead of Writing

1:00:31Humility and Feedback: The Iterative Process of Building Claude Code

Humility and Feedback: The Iterative Process of Building Claude Code

1:02:06Ticketing Systems and AI Coordination: The Evolution of Workflow

Ticketing Systems and AI Coordination: The Evolution of Workflow

Part 7: Expanding the Ecosystem: Cowork and Swarms

1:04:17Claude Cowork: Building a Product for Non-Engineers in 10 Days

Claude Cowork: Building a Product for Non-Engineers in 10 Days

1:07:10The Complexity of Cowork: Safety, Guardrails, and the Virtual Machine

The Complexity of Cowork: Safety, Guardrails, and the Virtual Machine

1:08:47Chrome Extensions and Project Management: Cowork in Action

Chrome Extensions and Project Management: Cowork in Action

1:10:11Electron and TypeScript: The Tech Stack Behind Claude Cowork

Electron and TypeScript: The Tech Stack Behind Claude Cowork

1:11:21Observability and Feature Flags: Rolling Out and Getting Feedback

Observability and Feature Flags: Rolling Out and Getting Feedback

1:13:43Agent Teams: Uncorrelated Context Windows and the Power of Swarms

Agent Teams: Uncorrelated Context Windows and the Power of Swarms

1:16:02Use Cases for Swarms: Building Plugins and Complex Features

Use Cases for Swarms: Building Plugins and Complex Features

Part 8: The Future of Engineering and Identity

1:17:23Embracing Change: Debugging Memory Leaks and the Speed of Progress

Embracing Change: Debugging Memory Leaks and the Speed of Progress