AI generating articles and writing code may already feel like yesterday’s story. The technology has now advanced to the point where it is taking on a significant portion of the thinking and research work within companies. Anthropic, the company behind Claude AI, has revealed that AI is now responsible for about 26 per cent of its AI research and development (R&D) work. That marks a dramatic increase from less than 1 per cent in February, highlighting how quickly the technology is progressing. Tasks that AI could barely manage a few months ago are now being handled with much greater responsibility.
The company’s latest data also indicates that more than 90 per cent of its AI research involves AI systems either working alongside humans or handling more advanced responsibilities.
The possibility of AI creating another AI, or even developing a more capable version of itself, has fascinated people since the early days of the technology. But this is no longer merely a theoretical possibility — it is already taking place.
AI agents are working inside Anthropic’s farm
The extent to which Anthropic is using AI internally is also notable. The company says approximately 30,000 AI agents are engaged in research and engineering through its most widely used internal platform at any given time.
Managing such a large number of agents requires extensive monitoring. Anthropic uses systems designed to track their activities. One monitoring system flags roughly 100,000 agent transcripts each week, although only around 50 are ultimately escalated to humans as high-priority cases. Another real-time monitoring system blocked just 0.002 per cent of more than one billion agent decisions made in August.
Safety gets a small share of the compute
AI safety remains a major subject whenever discussions turn to the rapid development of artificial intelligence. Anthropic’s data on compute allocation between July 13 and 20 showed that approximately 6 per cent of the computing resources assigned to AI R&D were devoted to safety-related work. When the focus was limited to AI-driven R&D, the proportion rose to around 12 per cent.
These numbers require some context. Computing resources do not provide a complete measure of the amount of effort devoted to AI safety, and Anthropic says its estimates are conservative. Nevertheless, the figures provide an indication of how the company is distributing its resources between advancing AI systems and ensuring their safety.
Anthropic’s broader argument is that there is currently limited public visibility into the pace at which AI capabilities are developing within leading frontier laboratories. The company believes AI labs should regularly publish measurements, adopt shared methodologies and permit independent verification of their findings.
