Coinbase Aims for 50% AI-Generated Code

Coinbase CEO Brian Armstrong announced on Wednesday that he aims for 50% of the cryptocurrency exchange’s daily code to be generated with 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.

> ~40% of daily code written at Coinbase is AI-generated. I want to get it to >50% 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.
>
> — Brian Armstrong (@brian_armstrong) September 3, 2025

According to Armstrong, AI-generated code currently comprises approximately 40% of Coinbase’s daily production. The CEO has been vocal about increasing AI’s role within the firm, even admitting on a podcast last month that he had dismissed programmers who resisted using AI coding tools. He later characterized this decision as “heavy-handed” and noted it was unpopular with some employees.

Armstrong’s goal reflects both an internal initiative at Coinbase and a wider movement within the tech industry. A report from OpsLevel revealed that by June 2025, 94% of tech firms had employees utilizing AI coding assistants, with increased productivity and quicker time-to-market cited as primary advantages.

AI-assisted coding typically involves developers utilizing AI for quick code generation while still engaging in debugging, reviewing, and maintaining a grasp of the AI’s operations.

However, this practice exists alongside the emerging trend of “vibe coding,” a term coined by computer scientist Andrej Karpathy, who previously served as a senior AI director at Tesla. Vibe coding signifies a departure from active code oversight, wherein developers accept AI suggestions entirely without understanding them and allow projects to progress largely beyond 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,” Karpathy shared on Twitter.

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

Garry Tan, CEO of Y Combinator, disclosed that a quarter of the accelerator’s Winter 2025 cohort relied on AI for 95% of their coding.

AI and Software Development

The rapid integration of AI has raised apprehensions concerning its potential impact on the job market for developers. Critics worry that excessive reliance on these tools might diminish people’s understanding of their creations, leading to overlooked safety concerns.

Concerns also exist surrounding the data used to train these AI tools. Art Abal, co-founder of the community-owned data network Vana, expressed support for Coinbase’s direction while voicing apprehensions.

“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.

“Without a better system of data ownership, humans risk becoming nothing more than ‘data cows’ endlessly milked, never compensated,” Abal added.