Claude is now writing most of Anthropic’s code.

June 10, 2026

Anthropic just released new internal benchmarks detailing exactly how much of their software is currently generated by their own AI. The numbers represent a massive shift in how frontier technology is made. Claude now autonomously writes more than 80% of the company’s merged codebase.

We aren’t just talking about simple auto-complete functions or a helpful coding assistant finding typos. The data shows Claude handling complex, open-ended software architecture tasks completely on its own. And it’s not just executing quick fixes—the model is managing these sophisticated engineering workloads autonomously over multi-day time horizons.

So what’s changing? Pure engineering velocity. By offloading this immense volume of cognitive labor to Claude, Anthropic has dramatically accelerated its engineering output by a factor of eight compared to 2024. The human engineers are no longer writing the bulk of the code; they are directing the AI that does.

The Recursive Loop

Why does this matter? Because we are watching the start of recursive self-improvement happen in real-time.

For decades, computer scientists have theorized about the moment when an artificial intelligence becomes capable of building a better version of itself. Anthropic’s data demonstrates that we have reached a major inflection point in making that theory a reality.

Anthropic is effectively creating a feedback loop where the AI accelerates its own development cycle. This loop actively speeds up the pace of creating new frontier models. As Claude gets smarter, it writes better code faster, which in turn helps human researchers build an even smarter version of Claude.

The Security Catch

Here’s the interesting part: moving this fast changes the risk profile. While this leap in engineering speed is a monumental technical achievement, it simultaneously heightens the urgency for robust alignment and security protocols.

When an AI system is largely responsible for coding its own architecture, the margin for error shrinks. Ensuring these advanced systems remain secure and aligned with human intentions isn’t just an abstract academic exercise anymore—it’s an immediate, practical necessity.

Accelerating Civilization

The convergence of artificial intelligence, computing, and digital infrastructure is fundamentally altering the future of civilization. As AI begins to demonstrate true recursive self-improvement, the timelines for solving complex bottlenecks in robotics, advanced manufacturing, and energy storage will likely compress drastically. We are moving rapidly toward an era where human engineering no longer scales at human speed, but is instead constantly amplified by the very systems we created.