Coinbase CEO Brian Armstrong said amid an increase in reliance on machine-generated development, almost half of the exchange’s daily code output is currently written by artificial intelligence.
In a September 3rd post on X, Armstrong revealed that around 40% of the code generated by Coinbase already comes from AI tools, predicting that by October 2025 the figure will exceed 50%.
However, Armstrong revealed that AI-generated codes were not uniformly applied across exchanges, highlighting the importance of reviews and human surveillance. He said:
“Obviously, you need to review and understand that, and not every area of your business can use AI-generated code, but you should use it as responsibly as possible.”
Still, this number is ahead of the huge tech companies like Microsoft and Google, which are around 30% generated on the machine of code.
His comments followed previous remarks about the cocky pint podcast, and told Stripe co-founder John Collison that Coinbase engineers were instructed to adopt AI development tools within a week.
Armstrong added that employees who resisted delegation were rejected, indicating how they turned into Coinbase’s engineering strategy.
“Red flag”
Armstrong’s push quickly elicited mixed reactions from the cryptography and technology community. Security specialists have expressed concern that passing such a large portion of mission-critical code to AI could expose them to vulnerabilities.
Larry Liu, founder of Dango Decentralized Exchange, called this approach “a giant red flag for security-sensitive businesses.”
Crypto Investor Blueprint: 5-day course on bag holdings, insider frontrunning, and lost alpha
Adam Cochran, partner at Venture Firm Cinneamhain Ventures, also questioned the wisdom that is highly dependent on AI. He said:
“I don’t know if 40% of our code is written by AI” is something we want to hear from where we store our financial assets. ”
Cochran added that AI can help with coding, but remains proven on a large scale for financial infrastructure.
The concerns of these industry leaders are not unfounded as it has become clear that AI code can “miss the relevant context” with bugs. This kind of error can be expensive on a Coinbase-sized platform, holding over $420 billion in digital assets on behalf of users.
However, other industry leaders have defended Coinbase’s approach.
Tensor co-founder Richard Wu said critics underestimate the maturity of the AI coding process.
He argued that 90% of high-quality codes could be generated by AI within five years if they maintain strict practices such as code reviews, automated testing, and lint.
WU also compares the poor output of AI with junior engineer errors, suggesting that a structured system mitigates both.
It is mentioned in this article
