Coinbase CEO Brian Armstrong said Wednesday that he hopes to use artificial intelligence to generate 50% of the cryptocurrency exchange’s daily code by October.
“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,” Armstrong tweeted.
Approximately 40% of the daily code written in Coinbase is generated by AI. I would like to be >50% by October.
Obviously, it needs to be reviewed and understood, and not all areas of your business can use AI-generated code. However, you should use it as responsibly as possible. pic.twitter.com/nmnsdxgosp
– Brian Armstrong (@brian_armstrong) September 3, 2025
According to Armstrong, AI-generated code already accounts for around 40% of Coinbase’s daily production. The CEO is frankly speaking with a push to expand the role of AI, admitting last month that he fired a programmer who resisted using AI coding tools. He later admitted that the move was “forced” and unpopular among some staff.
Armstrong’s target reflects both Coinbase’s internal push and a broader industrial shift. OpsLevel, which creates an internal development portal, found that as of June 2025, 94% of high-tech companies employ staff who employ AI coding assistants, with productivity and speed being cited as important benefits.
AI-assisted coding generally involves developers leaning on AI for quick code generation while debugging, reviewing and maintaining what a machine is doing.
However, it grew with the rise of “Vibe Coding.” This is a term popularized this year by Andrej Karpathy, former Tesla AI director and founder of Eureka Labs.
In his explanation, vibe coding means abandoning aggressive monitoring of the code completely, accepting wholesale AI proposals, pasting error messages without understanding them, and evolving the project outside of human understanding.
“Sometimes LLMS can’t fix a bug, so I’ll either get around it or ask for random changes until it’s gone. That’s not too bad for a disposable weekend project, but it’s still pretty entertaining,” he tweeted.
“I’m building projects and WebApps, but it’s not actually coding. I’m just looking at things, saying things, running things, copying pasting.
Garry Tan, CEO of Y Combinator, revealed that a quarter of the 2025 winter batch relies on AI for 95% of the code.
AI and Software Development
The onslaught of AI has led to concerns about the impact on the developer’s job market. There have also been concerns about whether safety issues could be overlooked as to whether the constant use of tools can reduce people’s understanding of what they are creating.
It is also a concern if the data used to train these AI agents come from. said Art Abal, co-founder of community-owned data network VANA. Decryption He thought Coinbase was moving in the right direction, but he had a question.
“I can’t help but wonder how much of the code generated in that AI is trained with human-generated data and how much of the human-generated value flows back to them.
“We need a better system of data ownership,” Abal added. “If not, those who trained these models would be excluded from the innovation and efficiency gains they have made possible. Without that, humans are nothing more than ‘data cows’, never compensated, never compensated. ”
