Over 95% of Coinbase's Code Is Now Written With AI

Rob Witoff, Coinbase's head of platform, said that 95% to 100% of the company's code is now written by or with large language models (LLMs) — a sharp jump from roughly 40%, the figure Coinbase CEO Brian Armstrong publicly cited back in February 2026.

What Coinbase actually said

"Effectively, 100% of our employees are using AI on a daily basis here," Witoff said, adding: "Close to 100% of our code, probably somewhere between 95% and 100%, is written by or with LLMs today." He said the share has more than doubled over the past several months compared to the February estimate.

Not everywhere runs unsupervised

Witoff stressed that the level of automation varies sharply by area. Internal prototyping is nearly 100% automated, but critical areas remain far more cautious: "When we're writing core cryptography, we have industry-leading cryptographers that are meticulously researching and reviewing one line at a time," he explained. Testing and verification for vulnerabilities and mathematical accuracy are also still done manually.

How this connects to the layoffs

In May 2026, Coinbase cut roughly 700 employees — about 14% of its workforce — framing the move as a push toward startup-level agility, with AI as a foundational part of operations rather than a supplementary tool. The company says its AI agents now handle work equivalent to roughly 1,200 employees, and that two or three senior staffers now cover tasks that previously required 10 or more people. Internally, the company has floated an even more ambitious projection — the equivalent of 100,000 employees by 2030.

Similar AI-driven layoffs this year have hit Crypto.com, Block, Kraken, Gemini and others — and it isn't limited to crypto: earlier in July, Meta employees sued the company, alleging layoffs were carried out with direct involvement from AI assistants. More on that case: our full coverage here

What this means in practice

The "95-100%" figure doesn't mean AI writes all of Coinbase's code unsupervised — by the company's own description, this is code written "by or with" models, with varying levels of human review depending on how critical the area is. But the trend is unmistakable: the share of automated development at major crypto companies is growing faster than headcount structures can adjust — and that mismatch, more than the technology itself, is increasingly what's fueling labor disputes across the industry.