Introduction

In this in-depth conversation, Anthropic’s Boris Cherny, the creator of Claude Code, and Alex Albert, who leads Claude relations, dive into the origins, capabilities, and future of Claude Code—a revolutionary tool that is reshaping the way developers interact with code. Unlike traditional coding assistants that live exclusively inside integrated development environments (IDEs), Claude Code is an agentic coding assistant that operates directly in your terminal, fitting seamlessly into your existing workflow regardless of your tech stack or preferred tools.

This article unpacks their discussion, capturing every detail, story, and insight—from the tool’s inception as an internal productivity booster to its public release and growing community adoption. We explore the practicalities of getting started, the evolution of software engineering, the power of Claude Code’s agentic capabilities, integration with GitHub Actions, and the nuanced tips and best practices that enable developers to harness its full potential. Whether you’re curious about the technological breakthroughs behind Claude 4 models or eager to learn how to orchestrate AI agents in your daily coding, this comprehensive article leaves no stone unturned.

What Is Claude Code and How Did It Begin?

Boris Cherny begins by describing Claude Code as “a way to do agentic coding in the terminal.” The core philosophy behind the tool is to remove barriers to AI-assisted coding by integrating it into the most universal and flexible interface developers already use: the terminal.

“You don’t have to adopt new tools, you don’t have to use new IDEs, you don’t have to use a particular website or anything. It’s just agentic coding, and it works wherever you work.”

This approach was born from Anthropic engineers’ diverse preferences and workflows. Boris highlights the variety within the engineering team—some use Zed IDE, others VS Code, and some swear by Vim, “never gonna take my Vim from me, pry it from my cold, dead hands.” Recognizing this diversity, the team chose the terminal as the “lowest common denominator” that everyone interacts with, ensuring Claude Code’s accessibility and flexibility.

Alex Albert adds that the terminal, by virtue of its simplicity, also facilitates rapid iteration and development:

“It also just happens to be the simplest, and because it’s so simple, we get to iterate really fast. And that’s, in hindsight, it turned out to be a good thing.”

This simplicity was not necessarily the initial intention but became a significant advantage in evolving the product quickly.

Getting Started: Installing and Using Claude Code

For developers eager to try Claude Code, the onboarding process is straightforward. Boris explains:

“You just download it from NPM. It’s npm install -g @anthropic-ai/claude-code… You just need Node.js on your system, and a lot of people have that — and that’s it.”

Once installed, you simply type claude into any terminal (whether iTerm2, Apple Terminal, or even an SSH or TMUX session), and Claude Code launches an interactive session guiding you through setup and usage.

Alex emphasizes the flexibility:

“One of actually the top ways that people use Claude Code is within IDE terminals. So, for example, you can run Claude within the VS Code terminal, and instead of seeing file edits in the terminal, you’re going to see them nice and big and beautiful in the IDE itself.”

This tight coupling with IDE terminals enhances the developer experience by leveraging IDE features and signals, making Claude Code smarter and more effective while retaining the seamless terminal interface.

Community Response and Internal Adoption

Claude Code was released publicly in February and has since experienced explosive growth. Boris shares the internal debate prior to release:

“We were like, ‘Is this secret sauce? Are we sure we want to give it to people?’ Because this is the same tool everyone at Anthropic uses every day.”

The tipping point came when the internal daily active user (DAU) chart for Anthropic employees shot straight up for several days, signaling high engagement and productivity gains.

“We were like, ‘All right, this is crazy. This is a hit.’”

Following internal success, limited external pilots received overwhelmingly positive feedback, confirming the decision to open Claude Code to the wider developer community.

Boris underscores the importance of dogfooding:

“Claude Code is written using Claude Code. Almost all the coding Claude Code has been written and rewritten using Claude Code… When you use a product that has obviously been dogfooded, you can feel it.”

This ethos of internal usage ensures the product is polished, intuitive, and truly meets the needs of developers.

Who Is Claude Code For?

Alex candidly addresses the cost considerations and ideal users:

“Claude Code is pretty expensive. So, if you’re coding on the weekends, you can try it for a little bit… But if you want to use it for more serious work, it’ll cost you $50, $100, $200 bucks a month, something like this.”

The pricing depends on usage intensity, but the sweet spot is around $50 monthly for serious developers. Enterprises, especially those managing large codebases, find Claude Code particularly valuable because it requires no indexing or complex setup:

“You just run it, and it works out of the box for pretty much every big code base in any language.”

Alex also explains the integration with Claude Max, Anthropic’s subscription service:

“Claude Code is part of Claude Max. You pay for the Max subscription… and you get pretty much as much Claude Code as you want. Practically, you’re not going to hit any rate limits.”

This unified subscription model simplifies access and encourages more frequent usage without worrying about API costs.

The Claude Code Experience: Agentic Coding in Action

Diving into how Claude Code works, Boris paints a vivid picture of a new paradigm for coding:

“If you’ve only used coding assistants in IDEs before… it’s not this at all. It’s super, duper agentic.”

Unlike assistants that merely autocomplete lines or small blocks, Claude Code comprehends your query, explores the codebase, reads files, gathers context, and then makes comprehensive edits autonomously.

This represents a fundamental shift in software development workflows. Boris shares a personal anecdote reflecting on the evolution of programming itself:

“My grandpa was actually one of the first computer programmers in the Soviet Union back in the 1940s… He programmed using punch cards… I think about the programming language and the experience of using the programming language as evolving in lockstep.”

He traces the journey from punch cards to assembly to high-level languages like COBOL, FORTRAN, Java, and then interpreted languages like JavaScript and Python. Each stage brought new abstractions and tooling improvements, culminating now in a mode where:

“You don’t have to deal with punch cards or assembly or even code, you deal with prompts, and the model figures out the coding part.”

This shift from manual code writing to orchestrating AI-driven coding assistants heralds a new era in software engineering.

Claude 4 and the Next Generation of Coding Models

Until recently, Claude Code was powered by Claude 3.7 Sonnet, which was powerful but sometimes challenging to steer. Boris notes:

“3.7 was kind of a beast. It’s an amazing coding model, but man, it’s hard to steer.”

For example, when asking it to write tests, it would often misunderstand the intent, requiring multiple attempts to correct. However, with the advent of Claude 4 models powering Claude Code, the experience has dramatically improved:

“Claude is just much better at holding your instructions… They typically do what you want the first shot.”

The Opus variant of Claude 4, in particular, represents a leap in capability:

“It feels like this next level above Sonnet… I haven’t written a unit test in months because Opus just writes my tests, and almost every time, it’ll one shot it perfectly the first time.”

This improved reliability enables more hands-off workflows and unlocks new use cases, such as integrating Claude Code into GitHub Actions for automated background coding tasks.

Integrating Claude Code with GitHub Actions: AI as Your Background Programmer

One of the most exciting developments is the ability to run Claude Code in GitHub Actions, allowing the model to autonomously tackle issues and generate pull requests without direct user interaction.

Boris explains the setup process:

“You open up Claude in the terminal… run /install GitHub Action, and that’ll walk you through this install step… It installs the Claude app in your GitHub repo.”

Once installed, developers can simply @mention Claude in issues or pull requests to delegate tasks:

“A coworker will put up a pull request, and instead of asking them, ‘Hey, can you fix this thing?’ I’ll just say, ‘Hey @Claude, fix this thing,’ and it’ll fix it.”

Similarly, writing tests becomes effortless:

“Instead of asking, ‘Can you write tests?’ I always feel kind of guilty… I’ll just say, ‘Hey @Claude, write tests.’ It’s just not a thing anymore.”

Alex reflects on the significance:

“That is like an entirely new aspect of programming right there… where we can basically pull in your always on-demand programmer to go fix these issues for you… operating in the background.”

This development marks a shift toward treating AI models as active collaborators, akin to human teammates, fundamentally changing software engineering dynamics.

The Evolving Software Engineering Mindset

Boris addresses the mental shift developers must embrace as coding transitions from manual craftsmanship to orchestrating AI agents:

“Some people really love controlling the code, and if you’re used to handwriting code, now the industry is shifting to a place where you’re orchestrating agents that write your code. It’s more about reviewing code than handwriting code.”

While some complex tasks still require direct coding, the window for manual intervention is narrowing as models improve.

“I dread hand writing code now because Claude is just so good at it.”

He predicts that as capabilities advance, programming will increasingly involve managing and coordinating AI agents rather than typing every line manually.

Typical Workflows: Balancing Hands-Off and Hands-On Coding

Boris outlines his personal workflow, which balances two modes:

  1. Simple Tasks: Bug fixes, writing tests, or small features are often delegated to Claude directly via GitHub issues or in terminal sessions running in auto-accept mode. He explains:

    “I have a couple of Claudes running in parallel, a couple of checkouts of our code base… I’ll hit shift-enter to enter into auto accept mode, and then I’ll come back in a few minutes.”

  2. Complex Tasks: For more involved work, Boris runs Claude within his IDE terminal, interacting more closely. If Claude gets stuck or the generated code needs refinement, he manually edits the code to achieve the “last mile” of polish.

This spectrum of difficulty dictates how much human involvement is needed, and developers must calibrate their expectations to avoid frustration:

“Sometimes people try to use it for too much at once… it gets choked up and you’re not happy with the result.”

He highlights the importance of adapting to evolving model capabilities:

“Every time there’s a new release, the capability grows… you can ask it for a little bit more every time.”

This reinforces the need to reset intuitions regularly as AI coding assistants improve at a rapid pace.

Tips and Tricks for Power Users

Boris shares some valuable strategies gleaned from Anthropic engineers and the broader community:

1. Make a Plan Before Coding

Before diving into implementation, ask Claude to brainstorm and outline a plan. This aligns intentions and improves outcomes:

“Ask it to make a plan and run it by you first… ‘Here’s the problem I want to solve. Before you code, brainstorm some ideas and make me a list of ideas… don’t write any code yet.’”

You can then select preferred approaches or combine options before giving the go-ahead to code.

2. Use Extended Thinking

Claude Code can interleave tool usage with deep reasoning:

“Ask it to read files first, then pause, then think and brainstorm some ideas, and then code.”

This mimics human problem-solving where initial research informs ideation before execution. Internal benchmarks confirm this approach yields better results.

The Power of Claude.md: Memory and Shared Instructions

One of Claude Code’s standout features is the use of Claude.md files—markdown files that serve as memory and instruction repositories for Claude across projects and teams.

Boris explains the different types:

  • Project-level Claude.md: Placed in the root of your repository, automatically loaded when Claude starts there. It can include instructions, preferred coding styles, architectural notes, or frequently run bash commands.

  • Team-shared Claude.md: Checked into version control, shared across team members to maintain consistency.

  • Personal Claude.local.md: A local-only version ignored by version control, for personal preferences.

  • Global Claude.md: Stored in a .claude folder in your home directory for instructions shared across all projects (used less frequently).

  • Nested Claude.md: Placed in subdirectories to provide specific instructions relevant to particular parts of the codebase.

This layered approach enables a rich, context-aware experience where Claude adapts to different scopes and preferences without repeated manual input.

Additionally, Boris describes using an interactive memory mode:

“When I see Claude do something really good or really bad, I’ll hit the pound sign to enter memory mode and tell Claude, ‘Hey, you should memorize this.’”

This dynamic learning further customizes Claude’s behavior in your environment.

What’s Next for Claude Code?

Looking forward, Boris and Alex outline two main directions:

  1. Deeper Integration with Tools: Expanding Claude’s native understanding and control of more IDEs, terminals, and continuous integration systems to create a seamless, ubiquitous coding assistant experience.

  2. Simplifying Access to Easy Tasks: Enabling Claude to be invoked outside terminals or GitHub, such as tagging it in chat apps to fix issues or perform coding tasks without leaving your communication platform.

They emphasize the importance of ensuring these new interfaces feel natural and productive before broad release.

Conclusion

Claude Code represents a bold reimagining of software development—shifting from handcrafted coding to orchestrated agentic programming powered by state-of-the-art AI. From its terminal-first design born out of Anthropic’s diverse engineering culture to its integration with GitHub Actions and advanced memory capabilities via Claude.md, it offers a powerful, flexible, and deeply collaborative coding experience.

As Claude 4 models continue to improve, the mental shift toward managing AI collaborators rather than typing every line is already underway. Developers are empowered to plan thoughtfully, delegate routine tasks, and focus on higher-level problem solving. With ongoing enhancements and integrations on the horizon, Claude Code is poised to become a ubiquitous companion in developer workflows, unlocking unprecedented productivity and creativity.

Boris and Alex’s conversation offers a rare, detailed window into this transformative technology, packed with practical guidance, inspirational stories, and glimpses into the future of AI-assisted coding.

If you’re a developer curious about what comes after traditional IDE assistants, Claude Code offers a compelling glimpse into the future—where prompts replace punch cards, and AI agents become your most trusted programming partners.

To learn more and get started with Claude Code, visit Anthropic’s official page and explore best practices.