Coinbase, a top name in the crypto world, is pouring effort into artificial intelligence (AI). They’re training their staff and using new AI tools to speed up how they build software. This move helps them stay ahead in the fast-changing crypto business. Brian Armstrong, who leads Coinbase, has a big goal. He wants AI to write half of their code as early as next month.
Armstrong recently shared on X that AI already generates about 40% of the daily code at Coinbase. This bold step shows how much the company believes in new ideas. But they also move carefully to keep things secure and high-quality.
The CEO wrote, “Approximately 40% of daily code written at Coinbase is AI-generated. I want to get it to >50% by October.” He posted this news to his followers on Wednesday.
~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. pic.twitter.com/Nmnsdxgosp
— Brian Armstrong (@brian_armstrong)
He added that humans must check and understand the code. Not every part of the business can use AI-made code. “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,” he said.
Armstrong linked to an official Coinbase blog post in the same message. This post explains how the company uses AI tools to help its developers work better.
Coinbase Adopts New AI Tools
The Coinbase blog post, released last month, lays out their strategy. They use a system called DORA (DevOps Research and Assessment) to track their progress. This system measures things like how fast changes are made, how often new code is released, how many mistakes pop up, and how much AI is being used.
At first, they just looked at the percentage of code AI wrote. Now, they track “tokens,” which are like processing units for AI models. This gives them an early sign of how much engineers are adopting AI. Their main goal is to make all engineers “AI natives.” These are pros who use AI tools every day without letting quality slip.
Coinbase uses specific AI tools like Cursor, Copilot, and Claude Code. They’ve also built their own systems so more than 1,500 engineers can try out AI safely.
Cursor has become the top choice for most Coinbase developers since February. AI adoption has been quicker for things like user interfaces (frontend), less sensitive backend tasks, and testing. AI truly speeds up these jobs. However, AI helps less with core systems, cloud setups, or vital exchange parts. For these tougher areas, Coinbase suggests using AI more for planning and testing phases.
Innovating with Technology
A chart from Armstrong’s post shows how much AI-generated code has grown at Coinbase. Starting around 15% in April, it hit about 45% by August. The company expects AI to write more code than humans by the end of the year. This shift has led to big wins. For example, one engineer can now rebuild or create entirely new code systems in days, a job that used to take months.
Still, it’s not all smooth sailing. Coinbase admits that using more AI has led to more mistakes in their code. That’s because big AI models don’t always create perfect results. To fix this, they worked with their security and privacy teams. Together, they made a “repository sensitivity matrix.” This matrix sets tough rules for when and how AI can touch important code, keeping customer data safe.
Armstrong put it simply: “tools don’t throw errors, humans do.”
AI: An Armstrong Priority
Brian Armstrong has made AI a top priority. He spoke recently on the “Cheeky Pint” podcast with Stripe co-founder John Collison. Armstrong shared that he ordered everyone at Coinbase to get comfortable with AI tools. He gave them just one week to learn programs like GitHub Copilot and Cursor.
He explained his thinking: “I thought: AI is important, we need everyone to learn it at least in theory. They don’t have to use it every day until we do some training, but they should at least know it in theory by the end of the week. If not, I’m going to organize a meeting with everyone who hasn’t done it and I’d like to meet with you to understand why.” This story was also reported by The Block.
Armstrong recalled that some employees didn’t follow through. Those who had no good reason for not learning were let go. While some workers found this move harsh, Armstrong stood by his decision. He said it gave the company the clear direction needed to truly embrace and learn the new technology. Even Armstrong admitted, “By the way, even as CEO, I use it a lot.”
A Pioneer in Crypto AI
Looking ahead, Coinbase plans to make its AI tools even better. They aim to fix slow spots in code reviews, testing, and dealing with tech problems. The company is also trying out AI “agents.” These smart tools could allow people who aren’t engineers to build automated solutions. They are getting ready for more advanced AI models that can think even better.
This aggressive AI strategy sets Coinbase apart. It makes them a leader in bringing AI into the world of major crypto exchanges. They are carefully balancing speed and new ideas with being responsible.