Artificial intelligence now helps write between 95% and 100% of Coinbase’s code, according to Rob Witoff, the crypto exchange’s head of platform. Cointelegraph reported that effectively all employees use AI daily, marking a rapid expansion from Coinbase’s estimate earlier this year that AI generated about 40% of its code.
Witoff said most engineers operate five to 10 AI agents at the same time. Those systems can research problems, write software, test changes, review code, prepare documentation, and complete other development tasks that previously required larger teams and longer production cycles.
Coinbase estimates that its active agents collectively perform coding work equivalent to roughly 1,200 employees. Witoff expects that figure to expand dramatically and has suggested that AI systems could handle the workload of 100,000 workers by 2030 if the technology continues improving at its current pace.
AI Agents Reshape Engineering Teams
The company is moving away from a model in which one engineer uses a single coding assistant. Instead, employees increasingly act as managers of multiple specialized agents, assigning tasks, reviewing outputs, resolving conflicts, and deciding which changes should move into production.
This approach can increase output, but it does not eliminate the need for human oversight. Cryptocurrency exchanges operate high-risk financial infrastructure where software errors can affect customer assets, trading, custody, compliance, and cybersecurity. Engineers remain responsible for validating AI-generated code and ensuring that agents do not introduce vulnerabilities or incorrect assumptions.
Coinbase has developed internal systems to evaluate AI models and assign work based on their strengths. Different models may be better suited to coding, research, data analysis, security review, or user-interface tasks. The company can route jobs between them rather than relying on a single provider.
Workforce Cuts Reinforce the AI Strategy
Coinbase cut approximately 700 employees, or 14% of its workforce, in May. Chief Executive Brian Armstrong said AI had dramatically changed the pace of work and argued that the company needed to return to the speed and focus of a startup with AI at the center of its operating model.
The restructuring reduced management layers and emphasized smaller teams led by experienced employees who can use AI tools effectively. Coinbase has also explored one-person teams in which an individual supported by agents covers responsibilities that once required separate engineering, product, and design roles.
The relationship between AI adoption and employment remains controversial. Companies often present automation as a productivity tool, but rapid increases in output can reduce demand for junior developers and other roles built around repetitive or well-defined tasks. At the same time, new positions may emerge around model evaluation, agent orchestration, security, and AI infrastructure.
Coinbase’s claims offer one of the clearest examples of an established financial technology company reorganizing its core development process around autonomous agents. The productivity gains could help the exchange launch products faster and operate with lower costs, but the model will be tested by code quality, system reliability, and the ability to manage increasingly independent AI tools.
If Witoff’s forecast proves accurate, software engineering may shift from writing most code manually to supervising large networks of agents. Coinbase is positioning itself as an early adopter of that transition, even as the workforce consequences become more visible.
