It is hard to find programmers these days who are not using AI coding assistants in some capacity, especially to write repetitive and mundane bits of code. However, those who refused to try the tools at Coinbase faced immediate consequences. After the company bought enterprise licenses for GitHub Copilot and Cursor, CEO Brian Armstrong said on the “Cheeky Pint” podcast that employees who declined to use the tools were promptly fired.
Armstrong explained that after securing licenses for every engineer, some at the cryptocurrency exchange warned him that adoption would be slow. They predicted it would take months to get even half of the engineers using AI. Shocked by this, Armstrong decided to take action. He posted a mandate in the company’s main engineering Slack channel. He stated that AI was important and required everyone to learn it and at least complete the onboarding process by the end of the week. He announced a meeting on Saturday for anyone who had not done so, saying he wanted to understand why.
At the Saturday meeting, Armstrong said some people had reasonable explanations for not setting up their accounts, such as being on vacation. However, some individuals did not have a good reason, and those employees were fired. Armstrong admitted this was a heavy-handed approach and acknowledged that some people in the company did not like it.
While it seems very few people were actually fired, Armstrong said the action sent a clear message that using AI is not optional at Coinbase. The entire story is striking, both for the fact that some engineers would not spend a few minutes trying the most hyped tech for coders and that the CEO was willing to fire them over it. Coinbase did not respond to a request for comment.
Since the incident, Armstrong has leaned further into AI training. The company now hosts monthly meetings where teams who have mastered creative ways to use AI share what they have learned.
Interestingly, podcast host John Collison, who has been programming since childhood, questioned how much companies should rely on AI-generated code. He commented that while it is clear AI is helpful for writing code, it is not clear how you run a codebase written by AI. Armstrong agreed with this sentiment.
This concern is echoed elsewhere in the industry. As previously reported, a former OpenAI engineer described that company’s central code repository as a bit of a dumping ground. The engineer said management had begun dedicating engineering resources to improve the situation.