There comes a time when a technology begins to help build itself. Anthropic has just published data suggesting that this moment has already begun.
In March 2024, Claude 3 Opus could autonomously complete software tasks taking around four minutes. A year later, the 3.7 Sonnet model leaped to tasks of an hour and a half. Today, the Opus 4.6 model already manages up to 12 hours of continuous work.
If this curve continues, tasks taking days could fall within the machine's grasp yet this year. And tasks taking weeks might be delegated to AI perhaps as early as 2027.
The impact of this autonomy has changed engineering from within. More than 80% of all code entering Anthropic's main codebase today is written by Claude itself. For comparison, before the start of 2025, this number was in the single digits. Because of this shift, the typical engineer at the company now approves and ships 8 times more code per day than they used to.
And this doesn't happen because they work longer hours. It happens because Claude writes the code, tests the system, finds bugs, and fixes issues autonomously, while the human simply assumes the role of a reviewer.
Anthropic calls this recursive self-improvement. In practice, it means an artificial intelligence helping to build the next generation of itself.
They make it clear they haven't reached the final stage of this concept yet. But the newly published data shows the path is already being traveled quickly. The loop is starting to close.
This is not a mere optimistic projection for the future. It is internal data showing what is already happening behind the scenes at this very moment.
The big question that remains: If artificial intelligence is already helping build the next artificial intelligence, what exactly do humans control in this cycle?
The research team's answer is direct. The human differentiator has shifted to good taste, judgment capability, and strategic vision to decide which problems really matter and are worth solving.
You can learn more about this on Anthropic's official site: https://www.anthropic.com/institute/recursive-self-improvement





Comments (0)
Comments are moderated and if they violate our Terms and Conditions of use, the comment will be deleted. Persistence in violation will result in a ban of your account.