Claude Code Was a $2 Billion Side Project

A Project Nobody Asked For Changed the Company

In September 2024, an engineer named Boris Cherny joined Anthropic. He wasn't placed in an official product team. It was an experimental division that people inside Anthropic compared to Bell Labs. There, Cherny started one experiment. Give Claude access to the file system.

The result was unexpected. When he shared it internally, engineers swarmed on their own. No one forced them. No internal announcement was made. CEO Dario Amodei later asked, "Did you force engineers to use this? Why is everyone using it?" Cherny's answer was simple. "I just gave them access. Everyone voted with their feet."

About a year and a half later. That side project became a product with annual revenue of $2.5 billion, roughly **$ 3.4 billion** in Korean won. The reason Anthropic became the leader in the AI coding market, and why OpenAI is now playing catch-up, all started with this one engineer's experiment.

Serendipity in a Bell Labs-Style Laboratory

Anthropic's experimental division had a unique structure. Like AT&T's Bell Labs, it was an environment where researchers could freely choose their projects. The transistor, laser, and UNIX operating system all came out of Bell Labs. Not because management ordered "build this," but because researchers found and solved problems on their own.

Cherny's experiment was in the same context. At first, he created a playful tool that displayed in the terminal what music a colleague was listening to using Claude. But when he gave Claude access to the file system, entirely different possibilities opened up. AI could autonomously perform all processes of reading, modifying, executing, and debugging code.

This tool spread like wildfire inside Anthropic within days. Not just the engineering team, but even non-developer roles started using it. When the internal adoption rate shot up, management took notice. It was the moment a side project converted to an official product track.

Business growth graph and data dashboard

Cherny later reflected in an interview, "When everyone starts using it without being told, it means the tool was solving a real problem." This statement pierces the essence of product development. A product that doesn't need to convince users. A product that spreads on its own once you open access. Claude Code was that kind of tool.

$1 Billion in 6 Months, Then$ 3.4 Billion

The numbers tell a clear story. Claude Code was publicly released in the first half of 2025. Within 6 months of release, it crossed $1 billion (approximately$ 1.37 billion) in annualized revenue (ARR). And as of February 2026, that figure climbed to $2.5 billion (approximately$ 3.4 billion).

Comparing this growth rate makes the context clearer. Compared to ChatGPT achieving $1 billion in annual revenue one year after launch, Claude Code passed the same milestone in half the time. What's more surprising is that a significant portion of this revenue was created through word of mouth alone. Anthropic's first large-scale marketing campaign was a Super Bowl ad in February 2026.

Metric	Claude Code	ChatGPT
Launch time	First half of 2025	November 2022
$1B ARR milestone	6 months post-launch	About 12 months post-launch
Primary growth driver	Word of mouth, developer community	Viral, media exposure
Current ARR	$2.5 billion	Undisclosed (est. $5B+)

Claude Code accounts for more than half of Anthropic's total enterprise revenue. The company's flagship product became a coding agent, not a chatbot. This changed Anthropic's business structure itself. A company that started as an AI safety research organization has shifted its identity to a developer tools company.

Anthropic's Super Bowl ad in February 2026 can be understood in this context. After the Super Bowl ad, Claude users surged by 11%. But looking at Claude Code's revenue growth curve, it was already on trajectory before the ad. The Super Bowl ad didn't trigger growth—it announced growth that was already happening.

A Tool That Wrote 90% of Itself

The strangest fact about Claude Code is this. 90% of Claude Code's codebase was written by Claude Code itself. The tool created itself. This is not a marketing phrase but an official figure disclosed by Anthropic.

The concept of self-referential development itself is unfamiliar. "Bootstrapping," where a compiler compiles itself, is an old tradition in computer science. But AI tools writing their own code is a different dimension. Bootstrapping is a completed compiler replicating itself, but in Claude Code's case, the tool is expanding and improving its own functionality. Like Escher's drawing of a hand drawing itself, the boundary between the creating subject and created object becomes blurred.

Even expanding to Anthropic's entire engineering team, it's similar. 70-90% of code produced company-wide is written by Claude Code. The era of developers typing code line by line has already ended inside Anthropic.

Cherny himself is the most extreme case. He revealed that he hasn't written a single line of code himself since November 2025. He delegates all coding work to Claude Code and focuses on architecture design and direction. In a Fortune interview, he said, "I don't write code anymore. Claude Code does everything."

This workflow works because Claude Code is not simply a code completion tool. If existing GitHub Copilot or Cursor are suggestive tools, Claude Code is an agentic tool. Tell it the goal and it reads files, identifies patterns, suggests changes, and executes directly. The developer's role changes to pressing the approval button. Some users let Claude Code operate independently for over 45 minutes.

The difference between suggestive and agentic is not just a functional difference. It's a paradigm shift. Suggestive tools increase developer productivity. Agentic tools change the developer's role itself. The former speeds up typing. The latter makes typing unnecessary.

Real-World Evidence From Spotify

The case that best demonstrates Claude Code's power in practice, not theory, is Spotify. Spotify's Principal Architect Niklas Gustavsson is a developer with 30 years of experience. In a Bloomberg interview, he said, "Code was always central for 30 years. Claude Code completely flipped that."

Spotify built an internal coding agent called 'Honk' based on Claude Code. Engineers request code modifications in natural language from Slack. Honk writes code, tests it, and creates pull requests. Over 650 agent-generated pull requests are merged to production every month. Spotify's official announcement states they reduced engineering time by up to 90% in complex code migration tasks.

Two-thirds of Spotify employees use Claude Code. This figure far exceeds the adoption rate of competing products. And the most experienced senior engineers have not written a single line of code since December 2025, according to reports. TechCrunch reported this with the headline "Spotify's top developers have written 0 lines of code since December."

This phenomenon is not unique to Spotify. Claude Code users work with this tool an average of 20 hours per week. That's half of a 40-hour work week. One Google engineer reported that Claude Code reproduced his one year of work in one hour. These testimonies are becoming a pattern, not individual cases.

Cases from users without technical backgrounds are also noteworthy. A farmer who created a tomato growing optimization app, a cafe owner who built a reservation system for local small businesses. Claude Code is becoming a tool that can convert ideas directly into software without needing to learn coding. This is the real meaning of vibe coding. An era where you can create software without being able to read code.

The Reversal Where OpenAI Is Playing Catch-Up

Before Claude Code, the AI coding market had a clear structure. Microsoft led with GitHub Copilot, and OpenAI supplied the core engine for that technology. Anthropic was known only for its chatbot Claude and had no presence in the coding tools market.

Claude Code flipped that board. Suddenly OpenAI had to chase Anthropic. It was the exact opposite structure. In the AI industry, Anthropic was always "second place." In the shadow of OpenAI, which had more capital, more users, and a wider ecosystem. Claude Code reversed that hierarchy at once. OpenAI rushed to upgrade Codex and even adopted specialized chips from Cerebras instead of Nvidia to respond. But the market flow had already tilted.

Engineering teams at Fortune 500 companies started adopting Claude Code. Even ordinary people without technical backgrounds used it to make apps. A new term emerged: 'vibe coding.' A method where you say what you want in natural language and AI generates the code. Claude Code was at the center of this vibe coding trend.

The competitors' responses can be summarized as follows:

Company	Product	Response Approach
OpenAI	Codex	Specialized chip adoption, open protocol
Microsoft	GitHub Copilot	Agentic feature enhancement, price cuts
Google	Gemini Code Assist	Gemini 3.1-based coding agent launch
Cursor	Cursor Editor	IDE-integrated agent advancement

The fact that the trigger for all this movement was one engineer's side project is noteworthy. Anthropic's corporate valuation reaches $380 billion. In the most recent funding round, they raised$ 30 billion. A significant portion of this astronomical figure is justified by Claude Code.

On social media, the expression "Claude-pilled" is trending. Meaning once you try it, you can't go back to other tools. This meme, which emerged spontaneously from the developer community, became more effective marketing than a Super Bowl ad. The time it took for a side project to become a company's core business and change the industry landscape was less than two years.

"By Year's End, the Title Software Engineer Will Disappear"

Cherny made a bolder prediction in a Fortune interview. "By the end of this year, everyone will become product managers doing coding. The title of software engineer will start to disappear. It will be painful for many people."

He compared this change to the invention of the printing press in the 15th century. Before Gutenberg's printing press appeared, scribes had the job of copying books letter by letter. When the printing press replaced that work, the scribe profession disappeared, but they moved to more creative areas like bookbinding, editing, and illustration. Cherny saw the same transition happening to software engineers.

Internal data at Anthropic supports this prediction. After adopting Claude Code, Anthropic engineers' productivity increased by 200%. Meaning the same number of developers are producing three times the code. This could be efficiency or grounds for restructuring. If the same work can be done with one-third the workforce, where do the remaining two-thirds go.

Cherny's advice was this. "Experiment with tools. Don't be afraid. Stay on the front lines." And he added that generalists skilled in multiple areas will have the greatest success. The logic is that people who can define problems and give correct instructions to AI become more valuable than experts in specific languages or frameworks.

Counterarguments to this prediction are significant. Software development is not just code writing. Requirements analysis, system design, incident response, security audits, legacy system maintenance. Code writing is only part of the total work. The counterargument is that AI writing code instead doesn't make engineers disappear—it changes the composition of engineers' work. In fact, Cherny himself still works as a Staff Engineer at Anthropic. He just doesn't write code, but his title hasn't changed.

An Era Where Side Projects Eat Industries

The most uncomfortable truth in Claude Code's story is this. A single side project can change the landscape of a multi-billion dollar industry. And the person who predicted the direction of that change was not the executive team, strategy team, or market analysts. It was the result of one engineer following their curiosity.

This pattern appears repeatedly in the tech industry. Gmail came from Google's 20% project. What started with one engineer's question "Why is email so inconvenient?" became a service used by 1 billion people. Slack was a byproduct from the gaming development failure of a company called Tiny Speck. The game failed but the internal communication tool survived and was acquired by Salesforce for $26 billion. Post-it Notes were born from a 3M adhesive experiment failure. Trying to make a strong adhesive resulted in a weak adhesive, and that 'failure' became a product with over$ 1 billion in annual revenue.

But Claude Code's case is decisively different in one way. Previous side projects 'opened' new markets. Gmail created the free webmail market, Slack created the enterprise messenger market. Claude Code is opening a new market while simultaneously threatening existing market participants. The profession of software development itself.

Anthropic is already preparing to expand beyond coding to healthcare, finance, and legal services. As Kate Jensen said, "code was just the first domain." What Claude Code proved is not the excellence of a specific tool. It's the possibility that AI agents can autonomously perform specialized knowledge work. Coding was just the first domino, and the next domino is already tilting.

Claude Code's success raises one question. What is Anthropic's greatest asset? AI model performance? Accumulated training data? No. An environment where engineers can freely experiment. Bell Labs created the transistor not because AT&T ordered "make a transistor." It was because they gave researchers time and freedom.

The moment Boris Cherny wrote a single line of code granting file system access in September 2024. At that time, he probably didn't know he was starting a $3.4 billion business. But historically, truly big innovations mostly started that way. Nobody asked for it, someone just did it. The problem is that this innovation is an opportunity for some and a cliff for others.