Claude (Quad Code)

Boris Churny joined Anthropic specifically for the mission-driven culture around AI safety. After a brief two-week detour to Cursor (where he realized the mission alignment mattered more than the product excitement), he returned to Anthropic determined to shape the future of AI. The vision was clear: Anthropic's path forward would be coding → tool use → computer use, mirroring their philosophy that more capable AI required deeper safety work.

Boris spent his first month at Anthropic hacking and prototyping weird ideas, most never shipped. Then he spent a month in post-training to understand the research layer beneath the product. This unconventional approach—understanding the model deeply before building—shaped everything that followed. The initial "Quad CLI" prototype was crude but revealing: when given a bash tool, Claude figured out how to use it to answer questions Boris hadn't explicitly instructed it to solve. He recorded a demo, posted it internally, and got exactly two likes. No one believed a terminal-based IDE made sense.

He kept building anyway, shipping 10-30 pull requests daily, all generated by Claude. The product evolved to match Claude's improvement trajectory: available in terminal, iOS, Android, desktop app, IDE extensions, Slack, and GitHub. By May 2024, Boris predicted that "by the end of the year, you might not need an IDE to code anymore"—a statement that made the room audibly gasp. He was right. By November 2024, 100% of his code was written by Claude, unedited since that month.

Internally at Anthropic, adoption went vertical once the product launched. Ben Mann's suggestion to track DAU early proved prescient. Externally in February 2024, the launch was surprisingly quiet. Claude Code wasn't an immediate hit; it took months for the market to understand what it actually was. The form factor—terminal-based, alien, unfamiliar—required an open mind and learning curve. But word spread. Early adopters got it. Within a year, companies from small startups to FAANG were building with it. Boris made feedback loops tight: a Slack channel captured all user feedback, and he'd fix reported issues within minutes to five minutes, making users feel heard and encouraging more feedback.

The underfunding principle worked brilliantly. Starting with just Boris forced speed, the only competitive advantage in a crowded market. As the team grew, they kept the principle: underfund projects slightly, force creative use of Claude, encourage people to go faster. Results were staggering. Productivity per engineer at Anthropic increased 200% in pull requests—compared to Meta's typical annual gains of a few percentage points across hundreds of engineers. Within a year, Anthropic's engineering team grew 4x while individual productivity soared.

What also worked: treating Claude as a coworker. By November, Claude wasn't just writing code—it was reviewing code (100% of Anthropic PRs), proposing features by analyzing feedback and bug reports, and suggesting what to build next. One engineer pointed Claude at a memory leak instead of using traditional heap snapshot debugging, and Claude solved it faster by writing its own analysis tool.

The biggest shift: generalists outperformed specialists. Everyone on the Cloud Code team—product manager, engineering manager, designer, finance, data scientist—coded. Engineers who crossed disciplines (product + infrastructure, design + engineering, business sense + engineering) were strongest.

One year after launch, Claude Code is the most impactful product Anthropic has shipped. 4% of all public GitHub commits are now authored by Claude, with private repo adoption much higher. Daily active users doubled in the past month alone. Boris's growth continues unabated—he ships 10-30 pull requests daily, 100% Claude-generated, unedited.

The frontier has shifted. Coding is solved. The next phase is agents doing general work: product managers using Claude to read feedback, suggest features, and coordinate work; non-technical users automating boring tasks (Boris had Claude pay a parking ticket and manage project sync-ups). By year-end, Boris predicts the title "software engineer" will be replaced by "builder," and "everyone will be a product manager and everyone codes." The printing press analogy resonates: democratizing capability, disruptive, transformative. The work now is understanding what this unlocks and managing the pain of transition.