The company quantified this shift using a prototype index that catalogs various R&D tasks, rating each by its degree of automation. While Claude currently collaborates with human staff on roughly 90% of these projects, the AI is not yet operating with full autonomy. Instead, the assistant acts as a force multiplier within the company's technical infrastructure.

To manage this transition, Anthropic has implemented oversight systems to monitor agent activity. As of August, approximately 30,000 AI agents were performing research and engineering tasks simultaneously on the company’s primary internal platform. Beyond development speed, the lab is also tracking its resource allocation; in July, between 6% and 12% of its total computing power for AI-related R&D was dedicated exclusively to safety testing and evaluation.

These disclosures arrive amid mounting industry pressure to address existential risks. Leaders including Dario Amodei, Sam Altman, and Elon Musk have voiced collective concern regarding the need for controlled development. To address these anxieties, Anthropic intends to grant independent, third-party evaluators access to internal data, mirroring the oversight capabilities currently held by its own risk assessment teams.