Coinbase head of platform, Rob Witoff, said that the company estimates between 95% and 100% of its code is written by or with large language models. It is a figure that stood at 40% just five months ago in February. The jump represents one of the strongest public disclosures of AI adoption from a major publicly listed financial technology company.
Witoff described AI coding as effectively universal across the company, with employees using LLM-powered tools daily for drafting, refactoring, testing, reviewing, debugging, and generating boilerplate.
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.
Don’t Miss Out on Our $1,000 USDT Airdrop on ByBit
Armstrong’s Leaner Operating Model
The disclosure follows Coinbase’s May 2026 restructuring, which cut approximately 14% of the company’s workforce, or around 700 employees. CEO Brian Armstrong linked the change to AI, saying AI had “dramatically” changed how work gets done and that Coinbase needed to return to startup speed with AI at its core. Armstrong also stated that engineers were using AI to accomplish in days what previously required entire teams working for weeks.
Coinbase’s earlier estimate in February was that AI was involved in about 40% of its code, and the company says it has since moved to nearly all-code AI assistance in a matter of months.
Supporters argue that AI-assisted development could improve COIN’s operating leverage if a smaller engineering team can ship and maintain products at equivalent or greater velocity.
Under that scenario, margins could improve while development cycles become shorter. What remains unquantified is the long-term maintenance and security cost of running financial infrastructure increasingly developed with AI assistance, a variable that critics argue the industry has not yet stress-tested at scale.
Discover: The Best Token Presales
The 40% to 95% Coinbase Vibe Coding Shift
The term vibe coding describes developers accepting AI output with minimal scrutiny. Coinbase’s model, as Witoff described it, emphasizes supervised AI assistance rather than unreviewed production output: AI can be used for drafting, refactoring, testing, reviewing, debugging, and generating boilerplate, while engineers remain responsible for oversight and deployment in production environments.
That nuance is easy to lose in the headline number. Coinbase’s claim is that code is written by or with LLMs, with engineers retaining responsibility for oversight and deployment. The adoption of AI tools across the crypto industry continues to accelerate, while formal guidance and governance frameworks are still evolving.
Coinbase adoption curve, from experimental productivity tool to near-universal operating model in under a year, mirrors the pattern seen at AI-native startups, but at the scale of a regulated, publicly listed exchange.
The competitive implication for peers is not trivial: a crypto exchange that can prototype, iterate, and ship with fewer employees has a structural speed advantage in a market where product velocity directly correlates with user acquisition and trading volume.
The layoffs complicate the narrative. Connecting 700 job cuts to AI productivity gains is exactly the kind of framing that draws regulatory and political scrutiny. Today, Armstrong has leaned into it explicitly rather than softened the connection.
For now, the primary variable is whether Coinbase’s supervised AI-assisted model holds up at scale, and whether the company’s security and reliability record supports the productivity claims once audited under pressure.
Discover: The Best Crypto to Diversify Your Portfolio