Claude writes 80% of Anthropic's own code as humans struggle to review it
It is happening. Anthropic just admitted that their AI, Claude, is basically building itself now, churning out almost all of their production code. The machines are self-improving, and the human engineers are officially glorified proofreaders.
The shift started accelerating after the launch of Claude Code, transforming the company's repository from a human-crafted project into an AI-generated empire. By May 2026, over 80% of the code merged into Anthropic's codebase was written by Claude. This conservative figure excludes experimental scripts, tracking only the actual lines that made it into production.
This automated surge has supercharged developer productivity, leaving human staff gasping for air. An average engineer in mid-2026 merges about eight times more code daily than they did in 2024. Instead of writing lines, humans now spend their days staring at Claude's outputs, acting as high-stress bottlenecks in a system that moves way too fast for biological brains.
To break this review logjam, Anthropic had to deploy another instance of Claude as an automated code reviewer. This AI inspector is designed to catch bugs before they hit production, with tests showing it would have stopped a third of the historical glitches on claude.ai. It turns out the only way to audit a hyperactive AI coder is to hire another hyperactive AI to watch it.
The quality of this synthetic code has rapidly matured, moving from sub-par spaghetti to professional-grade work. By mid-2026, Claude's success rate in solving open-ended programming tasks without clear instructions spiked to 76%. It is no longer just autocomplete; it is a colleague that is rapidly outgrowing its mentors.
This recursive loop brings humanity one step closer to the singularity, where software designs and trains its own successor without human intervention. The speed of this transition has apparently spooked even its creators, who are now openly suggesting that the world might need a global kill-switch to pause AI development if things get too wild. The ultimate irony is that the people building the future are already looking for the exit doors.
Source: Anthropic
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.