According to reports from Coinbase, the company has dramatically increased its AI coding adoption, with 95-100% of code now written by or with large language models, up from just 40% five months ago in February. Coinbase platform head Rob Witoff revealed in an interview with Cointelegraph that 100% of employees across the company use AI every day, with the company achieving this level of integration in just a few months. This represents one of the strongest public disclosures of AI adoption from a major publicly listed financial technology company.
The AI coding surge follows Coinbase's May 2026 restructuring that cut approximately 14% of the company's workforce, or around 700 employees. CEO Brian Armstrong linked the change to AI, stating that AI had "greatly" changed the work pace and that Coinbase needed to "return to startup speed and focus with AI at the core." The layoffs mainly affected junior developer roles and roles in departments including marketing, legal, and customer support. According to Coinbase engineers, 2-3 people can now do work that previously took more than 10 people, with the company completing work equivalent to about 1,200 employee equivalents in programming.
Coinbase engineers generally run 5-10 AI agents in parallel and have achieved remarkable productivity gains. As per Coinbase platform head Rob Witoff's interview with Cointelegraph, the company's level of AI reliance varies across different types of code, with core cryptography relying more on humans than prototype development. The company uses AI extensively for testing and ensuring code runs as expected, with no vulnerabilities. Witoff predicts that by 2030, AI integration could reach a workload equivalent to 100,000 employees.
For sensitive infrastructures like cryptography and core security systems, Witoff acknowledged that human oversight remains central. Coinbase operates trading systems, custody infrastructure, wallets, compliance tools, and blockchain integrations where software errors carry direct financial and regulatory consequences. The company emphasizes supervised AI assistance rather than unreviewed production output, with engineers retaining responsibility for oversight and deployment in production environments.
