Artificial intelligence systems are now writing the majority of their own code, according to a recent report from Anthropic. The company says AI handles increasingly complex research tasks, while humans focus on deciding which problems to tackle in the first place. The finding marks a notable shift in how human oversight is structured inside advanced AI development.
What the Report Says About Autonomy
Anthropic’s analysis describes a situation where the AI generates and refines most of its own software instructions. Humans step in mainly to set priorities—choosing which challenges the system should work on. This changes the traditional dynamic where a person wrote every line of code or directly guided each research step. Instead, the AI takes over the building and investigating, leaving people to act more like strategists.
The report doesn’t specify the exact percentage of code that is now AI-generated, but the language is clear: most of it. That represents a large jump from just a few years ago, when AI was seen mostly as a tool to assist human programmers, not as the primary writer.
Complex Research Becomes More Autonomous
Beyond code, the report highlights that the same AI systems are taking on research tasks that are more involved than simple pattern matching or data sorting. Those tasks, the company says, now include experiments and analysis that once required a human researcher to design and carry out. The AI decides on methods, runs tests, and interprets results. People still define the goals and decide when the answers are good enough.
This progression raises questions about how much decision-making will remain in human hands as the technology improves. Anthropic’s report doesn’t answer those directly, but it draws a line: at least for now, the high-level choices—what problem to solve—stay with people.
The shift has implications for both software engineering and scientific research. Teams that use these systems could find themselves spending less time writing code or doing routine lab work, and more time defining where the AI should point its efforts next.
