Claude Code Creator Boris Cherny Reveals How He Built the Future of AI Coding
Boris Cherny didn't set out to build one of the most talked-about AI coding tools in the world. But today, Claude Code — the terminal-based agentic coding tool from Anthropic — is being used by engineers at Salesforce, NASA, Uber, Deloitte, and thousands of startups worldwide. And a staggering 90% of Claude Code's own codebase is written by Claude Code itself.
In a special episode of Y Combinator's Lightcone podcast, Boris Cherny, the creator of Claude Code and Head of Claude Code at Anthropic, sat down to share the full story — from the accidental origin of the tool to his philosophy on building AI products, the future of software engineering, and advice for founders in the age of AI.
This article breaks down everything Cherny revealed in that conversation, packed with insights every developer and founder needs to hear in 2026.
Who Is Boris Cherny? The Man Behind Claude Code
Boris Cherny joined Anthropic in September 2024 — not to build Claude Code, but to work on a team called "Labs." That team no longer exists because, as Cherny puts it, they "fulfilled their purpose." The purpose turned out to be something much bigger than originally imagined.
Before Anthropic, Cherny worked at Meta and other major tech companies. He's also well known in the TypeScript community — a parallel he draws himself when discussing the philosophy behind building Claude Code.
What makes Cherny's story interesting isn't just the tool he built. It's how he thinks about building in an era where the model itself is the product — and the model keeps getting better every few months.
How Claude Code Was Actually Born
Claude Code was not planned. It was discovered.
In the early days at Anthropic's Labs team, Cherny and his colleagues were experimenting with what AI models could do autonomously — running commands, editing files, working in a terminal. What started as an internal experiment quickly showed signs of something transformative.
"The most surprising moment," Cherny said in the podcast, "was realising just how capable the model already was — and how little scaffolding it actually needed."
That insight became the founding principle of Claude Code: elegant simplicity over complex tooling. Rather than wrapping the AI in an elaborate interface with menus, buttons, and click-through workflows, the team chose the terminal — a decades-old, minimal environment that experienced developers already knew and trusted.
The terminal, Cherny explained, has a kind of timeless elegance. It is direct, flexible, and composable. It forces clarity. And crucially, it gets out of the way of the model.
The Elegant Simplicity of the Terminal
One of the most fascinating parts of Cherny's interview is his defense of the terminal as the right home for an AI coding agent.
In an era where every app competes to have the most beautiful UI, Claude Code launched as a command-line tool. That was a deliberate and deeply considered choice.
"Designing for the terminal was hard," Cherny admitted. "But we kept coming back to the fact that the terminal is where developers live. It's composable, it's scriptable, and it doesn't get in the way."
The terminal also taught the team something important: verbosity matters. One of the key design questions was — how much should Claude say while it's working? Too little and users feel lost. Too much and it becomes noise. Getting that balance right, Cherny said, required constant iteration and a beginner's mindset.
What's in Boris Cherny's CLAUDE.md?
For those unfamiliar, CLAUDE.md is a special file that Claude Code reads to understand the context of a project — coding conventions, architecture decisions, style preferences, and more. Think of it as a briefing document you write for your AI teammate.
Cherny was surprisingly candid about what's in his own CLAUDE.md:
- Project-specific coding conventions
- Notes on architecture and folder structure
- Preferences around how Claude should communicate progress
- Context about the team's current priorities
The takeaway for developers: the better your CLAUDE.md, the better your results. Claude Code is not a magic button — it's a highly capable collaborator that performs best when given clear context upfront.
Beginner's Mindset: The Key to Building with Improving Models
One of Cherny's most repeated themes in the interview was the importance of approaching Claude Code — and AI tooling in general — with a beginner's mindset.
This might sound counterintuitive. Shouldn't experienced developers get the most out of Claude Code? Not necessarily, according to Cherny.
"As the models improve, a lot of the workarounds and hacks that experienced users developed become irrelevant — or worse, counterproductive," he explained. "People who come in fresh, without assumptions, often use the tool more effectively."
This is especially important now, in mid-2026, as new model versions arrive every few months. Techniques that worked six months ago may already be outdated. The best Claude Code users, Cherny suggested, are those willing to unlearn and re-experiment constantly.
Hyper-Specialists vs. Hyper-Generalists: Who Wins?
One of the most thought-provoking sections of the interview was Cherny's take on the specialist vs. generalist debate — particularly relevant for the future of software engineering roles.
His view: the hyper-generalist will have the edge in the AI era.
Why? Because Claude Code can handle much of the deep, specialist technical execution. What it can't replace is broad judgment — knowing which problem to solve, how to communicate with stakeholders, when to ship and when to hold, and how to navigate ambiguity.
"The most valuable people," Cherny said, "will be those who can move across problem spaces quickly — and use AI to go deep when they need to."
This connects directly to his broader prediction about the software engineering title itself. Cherny has publicly stated he believes the job title "software engineer" will begin to disappear — not because coding stops mattering, but because the boundary between engineer, product manager, and founder is dissolving. The new title might be "builder."
Subagents and the Future of Claude Code
Claude Code today is capable of far more than writing functions or fixing bugs. One of the most powerful features Cherny discussed is subagents — the ability for Claude Code to spin up parallel instances that tackle different parts of a problem simultaneously.
Imagine asking Claude Code to: refactor a module, write tests for it, update the documentation, and check for security issues — all at the same time, across separate subagent sessions. That's the direction things are heading.
Cherny himself runs 5 to 10 Claude sessions simultaneously, across his terminal and phone, each working on a different task. He kicks off a set of tasks in the morning from his phone before reaching the office, then reviews progress when he arrives.
"I haven't written a line of code by hand in about eight months," Cherny has said publicly. Every line of Claude Code is written by Claude Code.
Plan Mode: Should It Exist at All?
An interesting philosophical moment in the interview came when Cherny was asked about Plan Mode — the feature in Claude Code that lets the model think through a problem before taking action.
"A world without plan mode?" Cherny mused. "Maybe, one day."
The idea is that as models become more capable and reliable, the need for explicit planning steps may diminish. The model will simply know to think before acting — without the user having to invoke it. For now, Plan Mode remains one of the most valuable features for complex, multi-step tasks. But it reflects the team's mindset: build for the model six months from now, not the model today.
The Vision for Claude Teams
Claude Code isn't just for solo developers. Cherny has a clear vision for how it fits into team environments.
The Claude Teams product is designed to let multiple engineers collaborate with AI as a shared, context-aware resource — one that understands the codebase, the team's conventions, and the history of decisions made across a project.
This raises an interesting human challenge Cherny was honest about: when engineers talk to Claude more than to each other, team culture can suffer. At Anthropic, they've responded by intentionally scheduling peer programming sessions and social time — to ensure that human collaboration doesn't atrophy just because AI is handling more of the code.
"You have to feel very safe being wrong," Cherny said. "In an environment where the model changes everything fast, psychological safety matters more than ever."
Advice for Founders: Build for What's Coming
Cherny's advice for founders building with Claude Code — or building AI products in general — was direct and memorable:
"We don't build for the model of today. We build for the model six months from now."
If you optimize your product for today's model limitations, you'll be leapfrogged the moment the next model drops. Cherny's team has a framed copy of Rich Sutton's The Bitter Lesson on the wall in the Claude Code area — the key insight being: the more general model will always beat the more specific model. Never bet against the model.
For dev tool founders specifically, his advice was:
- Don't over-engineer scaffolding around the model — it becomes tech debt fast
- Invest in understanding the model's capabilities deeply, not just the API surface
- Think about distribution early — the best tool means nothing if developers don't discover it
- Build for trust: developers will only hand off real work to AI tools they genuinely trust
How Much Life Does the Terminal Still Have?
Given the rise of web-based AI coding tools and browser interfaces, one fair question is: how long will the terminal remain Claude Code's home?
Cherny's answer was confident: a long time.
And he's backed that up with action. Claude Code is now available on standard web browsers — not just command-line tools — making it more accessible to developers on phones and tablets. But the terminal remains the core experience, and Cherny believes that won't change soon.
"The terminal is where serious development happens," he said. "It's not going anywhere."
The Future of Coding: What Boris Cherny Actually Believes
Perhaps the most significant takeaway from the entire interview is Cherny's honest, measured take on where coding is heading.
He is not saying developers are obsolete. He is saying the work is changing — fundamentally and fast.
At Anthropic, between 70% and 90% of all code is AI-generated. For Claude Code specifically, around 90% of its own code is written by itself. At Y Combinator's latest batch, half of all founders reported that Claude Code writes 100% of their code. Out of a couple hundred founders, only one hand went up when asked if they write code without AI at all.
"Coding is getting solved for a bigger and bigger percentage of the code we write," Cherny said.
The engineers who thrive won't be those who resist this shift. They'll be those who embrace it — using AI to multiply their output, their scope, and their ambition.
Frequently Asked Questions (FAQ)
Who created Claude Code? Claude Code was created by Boris Cherny, who joined Anthropic in September 2024. He serves as Head of Claude Code at Anthropic.
What is Claude Code used for? Claude Code is an agentic AI coding tool that can write code, fix bugs, run terminal commands, update files, perform code reviews, and integrate with developer tools like Asana — all autonomously from the command line or browser.
Is Claude Code free? Claude Code is available on Claude Pro ($17/month) and Claude Max ($100/month) plans. It is also accessible via the Anthropic API for developers building on top of it.
What is CLAUDE.md? CLAUDE.md is a project-specific context file that Claude Code reads to understand your codebase, coding conventions, and preferences — helping it deliver better, more relevant results.
Will AI replace software engineers? According to Boris Cherny, the title of software engineer may change or fade, but the need to build software won't disappear. The work is shifting toward higher-level thinking, product judgment, and AI collaboration rather than hand-writing every line of code.
What is Plan Mode in Claude Code? Plan Mode is a feature that lets Claude Code think through a complex problem and outline its approach before executing — useful for multi-step or risky operations where you want to review the plan first.
Conclusion
Boris Cherny's story is one of the most fascinating in tech right now — a developer who accidentally built a tool that is redefining how the entire industry writes software, and who is candid, thoughtful, and principled about what that means.
Whether you're a developer trying to 10x your output, a founder building the next great product, or simply someone curious about where AI is taking us — the lessons from this interview are essential reading.
The future of coding isn't about writing less code. It's about building more, faster, and with far greater ambition than any single developer could achieve alone.
And Claude Code, it turns out, was just the beginning.
Sources: Y Combinator Lightcone Podcast — "Inside Claude Code With Its Creator Boris Cherny" (Feb 17, 2026); Fortune; PCMag; Final Round AI
Published on nexusblog.in | Tags: Claude Code, Boris Cherny, Anthropic, AI Coding Tools, Developer Tools, Y Combinator, Future of Software Engineering
Comments
Post a Comment