Artificial intelligence now assists with more than 95% of the code written at Coinbase, marking a sharp increase from around 40% earlier this year as the crypto exchange accelerates its adoption of AI across engineering and product development.
The figures were shared by Rob Witoff, Coinbase’s Head of Platform, who said nearly every employee at the company now uses AI in their daily work. Engineers typically work alongside five to 10 AI agents simultaneously, using them to generate code, review changes, identify bugs and automate repetitive development tasks.
The company stressed that AI is not replacing engineers entirely. Instead, developers remain responsible for reviewing, testing and approving code before it is deployed, particularly for security-critical systems such as cryptography and core infrastructure. According to Witoff, highly sensitive code still undergoes rigorous human review despite being developed with AI assistance.
Coinbase’s AI adoption has accelerated rapidly over the past few months. In September 2025, CEO Brian Armstrong said AI generated about 40% of the company’s code and set a target of reaching 50%. The latest figures suggest the company has moved well beyond that goal in less than a year.
The update follows Coinbase’s decision in May to reduce its workforce by about 14%, or roughly 700 employees. At the time, Armstrong told staff the company needed to return to “startup speed” by placing AI at the centre of its operations. The latest disclosure offers a clearer picture of how deeply AI has become embedded in the company’s software development process.
According to Witoff, AI has changed how engineering teams are organised. Smaller groups of experienced developers can now oversee work that previously required much larger teams, allowing engineers to focus more on architecture, product decisions and quality assurance while AI handles much of the routine coding.
Coinbase’s approach reflects a broader shift across the technology industry, where companies are increasingly relying on large language models to speed up software development. AI coding assistants are now commonly used for writing, debugging, testing and documenting software, although many organisations continue to require human oversight for security-sensitive and production systems.
For Coinbase, the latest milestone is less about replacing software engineers and more about changing how they work. As AI takes on a larger share of coding tasks, developers are increasingly acting as reviewers, architects and decision-makers, guiding AI-generated output rather than writing every line of code themselves.