Brian Armstrong Wants Half of Coinbase Code To Be AI-Written By October

In brief

Coinbase CEO Brian Armstrong wants half of all code at the exchange to be AI-generated by October.
AI already accounts for 40% of daily code at Coinbase, reflecting a wider industry embrace of AI coding assistants.
The surge in AI-assisted coding contrasts with the “vibe coding” trend, where developers lean on AI for coding with minimal oversight.

Coinbase CEO Brian Armstrong said Wednesday that he wants 50% of the cryptocurrency exchange’s daily code to be generated using artificial intelligence by October.

“Obviously it needs to be reviewed and understood, and not all areas of the business can use AI-generated code. But we should be using it responsibly as much as we possibly can,” Armstrong tweeted.

AI-generated code already makes up roughly 40% of Coinbase’s daily production, according to Armstrong. The CEO has been outspoken in his push to expand AI’s role across the company, even acknowledging on a podcast last month that he had fired programmers who resisted using AI coding tools. He later admitted the move was “heavy-handed” and unpopular among some staff.

Armstrong’s target reflects both Coinbase’s internal push and a broader industry shift. OpsLevel, which creates internal dev portals, found that as of June 2025, 94% of tech companies had staff employing AI coding assistants, with productivity and speed to market cited as the key benefits.

AI-assisted coding generally involves developers leaning on AI for rapid code generation while still debugging, reviewing and maintaining an understanding of what the machine is doing.

But it has also grown alongside the rise of “vibe coding,” a term popularized this year by computer scientist Andrej Karpathy, former senior AI director at Tesla and founder of Eureka Labs.

In his description, vibe coding means abandoning active oversight of the code entirely and accepting AI suggestions wholesale, pasting in error messages without understanding them, and letting projects evolve largely outside human comprehension.

“Sometimes the LLMs can’t fix a bug so I just work around it or ask for random changes until it goes away. It’s not too bad for throwaway weekend projects, but still quite amusing,” he tweeted.

“I’m building a project or webapp, but it’s not really coding – I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”

Y Combinator CEO Garry Tan also revealed a quarter of the accelerator’s Winter 2025 batch relied on AI for 95% of their code.

AI and software development

The onslaught of AI has led to concerns about its impact on the job market for developers. Concerns have also been raised over whether constant use of the tools could decrease people’s understanding of what they are creating—with the result that safety issues could be overlooked.

Where the data used to train these AI agents comes from is also a concern. Art Abal, co-founder of community-owned data network Vana, told Decrypt that while he thought Coinbase was moving in the right direction, he had questions.

“I can’t help but wonder how much of that AI-generated code was trained on human-created data, and how much of the value those humans created is flowing back to them? My guess is close to none,” he said.

“We need a better system of data ownership,” Abal added. “Otherwise, the very people whose data trained these models will be excluded from the innovation and efficiency gains they made possible. Without it, humans risk becoming nothing more than ‘data cows’ endlessly milked, never compensated.”