Key Takeaways
- Coinbase has started measuring the percentage of code that is AI-generated.
- The company expects AI-generated code to “eclipse” human output by the end of the year.
- CEO Brian Armstrong said Coinbase should be using AI-generated code “as much as we possibly can.”
Brian Armstrong said that he wants more than half the code written each day at Coinbase to be AI-generated by October.
Armstrong’s statement reflects Coinbase’s increasingly AI-centric approach to software development, which has seen the company adopt Large Language Model (LLM)-powered tools like Cursor, Copilot, and Claude Code.
Coinbase Measuring Percentage of Code Written by AI
In an X post on Wednesday, September 3, Armstrong shared a graph that shows the percentage of Coinbase’s daily code output that is generated by AI.
From less than 20% at the start of April, the figure has climbed to more than 40% today, and Armstrong said he wants to reach more than 50% by October.
Coinbase’s Kyle Cesmat and Chitra Venkatramani echoed that sentiment in a blog post, writing that “AI is on track to eclipse human-generated code at Coinbase by the end of the year.”
Turbocharging Developer Efficiency
Coinbase isn’t alone in its efforts to streamline previously labor-intensive coding tasks using AI.
Across the software industry, developers are using AI code assistants to automate workflows. Such tools are especially useful for tasks that are well-defined and repetitive, like generating standard database operations or API client code.
“This has enabled profound success stories that weren’t possible 12 months ago, like single engineers refactoring, upgrading or building new codebases in days instead of months,” Cesmet and Venkatramani observed.
Armstrong stressed that new AI-augmented workflows don’t make human developers obsolete. Code still needs to be reviewed and understood, “and not all areas of the business can use AI-generated code,” he said.
Nonetheless, “we should be using it responsibly as much as we possibly can,” the Coinbase CEO concluded.
AI “Not a Magic Bullet”
Exploring where AI has had the most impact, Cesmet and Venkatramani found that teams working on front-end features were adopting LLMs at a faster rate.
Meanwhile, those working with low-level systems and sensitive data, where AI hallucinations can result in security vulnerabilities, have not experienced the same “meaningful increase” in AI usage.
“Leveraging LLMs for coding is not a magic-bullet we should expect teams to universally adopt.”
“Our most senior engineers might spend weeks finding the right fixes to make, and it’s important to focus on results, not any single path to get there,” they emphasized.
Recommended Secure Partners
James Morales is CCN’s blockchain and crypto policy reporter. He has been working in the news media since 2020, writing about topics such as payments, banking and financial technology. These days, he likes to explore the latest blockchain innovations and the evolving landscape of global crypto regulation.
With an educational background in social anthropology and media studies, James uses his platform as a journalist to explore how new technologies work, why they matter and how they might shape our future.
