Coinbase's head of platform Rob Witoff said 95%-100% of the exchange's codebase is now written by or with large language models, up from 40% in February, according to an interview with the executive.
The shift represents a rapid adoption of AI-assisted development across one of the largest U.S. cryptocurrency platforms. Witoff did not specify which LLMs Coinbase uses or whether the figure includes all software across the organization or specific business units.
Coinbase has roughly 2,500 employees and operates the exchange, custody services, developer APIs, and a blockchain protocol. The company's engineering team writes and maintains code for trading systems, compliance infrastructure, wallet functionality, and mobile applications, among other systems. Witoff stated that generative AI has been integrated into the development workflow at significant scale in a five-month period.
OpenAI's GitHub Copilot and similar products have seen adoption in professional engineering environments. Coinbase did not clarify whether the percentage counts lines of code written without human authorship or code reviewed and modified by engineers after initial generation by AI.
The February 40% figure serves as the baseline for comparison. Witoff offered no detail on whether the increase stems from new hiring of AI-capable engineers, mandatory adoption of LLM tools across teams, or organic adoption by developers.
Coinbase has not announced a formal AI strategy statement or released engineering guidelines on LLM use in production systems. The company did not respond to requests for clarification on the scope or methodology of the measurement.
