The cryptocurrency exchange Coinbase has reached an impressive milestone in development automation: neural networks now generate over 95% of the platform's entire software code. This figure has more than doubled since February, when the AI's share in code writing was estimated at 40%. According to Coinbase CEO Rob Witoff, artificial intelligence is now used daily by 100% of the company's employees.
The degree of automation varies depending on the complexity of the tasks. Prototyping is entirely delegated to neural networks—here AI operates without human involvement. Core systems are developed using a hybrid model, where algorithms and engineers work in tandem. However, the critically important area—cryptography—remains under the full control of specialists. Every line of code here is manually verified, and AI only helps identify vulnerabilities and verify mathematical calculations.
Restructuring and Workforce Optimization
In May, Coinbase conducted a large-scale staff reduction, laying off 700 people—14% of its workforce. CEO Brian Armstrong explained this step as necessary to restore the company's "startup speed and focus" through the active implementation of AI. The layoffs primarily affected junior developers, as well as employees in marketing, support, and compliance.
The adoption of technology has radically changed team structures. Now, groups of 2-3 senior specialists successfully handle tasks that previously required the involvement of 10 or more people. Each Coinbase engineer currently has between 5 and 10 active AI agents. Collectively, these virtual assistants perform a volume of work equivalent to that of 1,200 employees.
Future Projections
Coinbase expects that by 2030, the number of virtual workers will grow to 100,000. This means automation will become not just a tool, but the foundation of the exchange's operational model. The trend of replacing humans with neural networks is also observed among other market players: in April, the Snap team announced the reduction of 1,000 employees to save $500 million, and in May, Dune announced a restructuring with the layoff of 25% of its staff.
My expert commentary: Coinbase demonstrates that implementing AI in development is not just hype, but a real path to increasing efficiency. However, full code automation carries risks: neural networks can generate errors that junior developers used to catch at early stages. By keeping cryptography "manual," the exchange is correctly prioritizing, but the question remains how sustainable such a model will be when scaled to 100,000 agents.