Coinbase's head of platform, Rob Witoff, said this week that between 95% and 100% of the company's code is now written by or with large language models. That's up from 40% in February 2026. The shift reflects a rapid adoption of AI-assisted coding at a regulated, publicly listed exchange.
From 40% to near-total adoption in five months
Witoff stated the AI-assisted coding covers drafting, refactoring, testing, reviewing, debugging, and generating boilerplate. Human oversight remains for sensitive areas like cryptography and core security. The company's adoption curve — from experimental tool to near-universal operating model in under a year — mirrors AI-native startups, but at a different scale.
The restructuring link
In May 2026, Coinbase cut about 14% of its workforce — roughly 700 employees. CEO Brian Armstrong linked the layoffs to AI-driven productivity gains. He said engineers using AI can accomplish in days what previously required entire teams working for weeks. The timing isn't coincidental.
Vibe coding vs. supervised assistance
The term 'vibe coding' describes developers accepting AI output with minimal scrutiny. Coinbase's model emphasizes supervised AI assistance rather than unreviewed production output. Still, the long-term maintenance and security costs of AI-developed code remain unquantified. Critics argue the industry hasn't stress-tested this at scale.
The next question is whether other major exchanges follow Coinbase's lead — and what happens when the first critical bug in AI-generated code surfaces in production. No one has an answer yet.
