Did Coinbase CEO Brian Armstrong’s AI mandate open Pandora’s box?
When Coinbase CEO Brian Armstrong went “rogue” and gave engineers at the crypto trading platform an ultimatum to use AI to write code or lose their jobs, he may have set a precedent for other CEOs. But in the process, he may have just let go of some of the most critical and creative coders in his industry. We look at the two camps of engineers on the AI divide and why companies probably need both.
The Coinbase mandate
Coinbase CEO Brian Armstrong required all engineers to adopt AI tools like GitHub Copilot and Cursor within one week after the company purchased enterprise licenses.
I went rogue. I posted in the all-in Slack channel… I was like: ‘AI it’s important. We need you to all learn it, at least on board. You don’t have to use it every day yet until we do some training, but at least on board by the end of the week.’
–Brian Armstrong, Coinbase CEO
The consequences
He held a meeting on a Saturday with the engineers who had not complied. Those who had valid excuses (e.g., being on vacation) were not penalised, but those who did not were fired.
The justification
Armstrong admitted in a podcast interview with John Collison, the co-founder of Stripe, that his approach was “heavy-handed” but believed it was necessary to signal that AI adoption is a non-negotiable priority for the company.
“I went rogue. I posted in the all-in Slack channel… I was like: ‘AI, it’s important. We need you to all learn it, at least on board. You don’t have to use it every day yet until we do some training, but at least on board by the end of the week.’”
He admitted that some of them “didn’t [have a good reason] and they got fired.”
“Some people really didn’t like it, by the way, that heavy-handed approach. But I think it did set some clarity, at least that we need to lean into this and learn about it.”
So we’re testing the limits of it. You know, when can it actually start to be the decision maker on some things and do better than humans?
–Brian Armstrong, Coinbase CEO
Company strategy: less human coding
Coinbase is now hosting monthly AI training sessions where teams share how they are using AI creatively. Armstrong has an ambitious goal for AI to write 50% of the company’s code by the end of the quarter.
“Even as CEO, by the way, I use it a lot,” justified Armstrong in the interview. We have a row now for AI that writes its input in as one of the people who help make decisions.”
He continued, “So we’re testing the limits of it. You know, when can it actually start to be the decision maker on some things and do better than humans?”
Wider context
Armstrong’s stance reflects a broader industry trend where CEOs are increasingly making AI adoption a strategic priority, though it has received mixed reactions from engineers who code, employees and industry observers.
Collison pointed out in the Armstrong interview, “It’s clear that it is very helpful to have AI helping you write code. It’s not clear how you run an AI-coded base.”
The AI coding debate: which side are you on?
According to UK-based Kyle Redelinghuys, an independent engineer who helps businesses looking to combine AI, fintech and practical software development, “AI for coding is not meant to replace developers but to amplify their productivity.”
He put these two types of developers into camps (and he’s in one of them):
Craft-focused developers are the ones who enjoy the process of writing code. They see programming as creative expression and worry that AI code writers remove the intellectually satisfying parts. For them, GitHub Copilot or Cursor feels like having someone else solve crossword puzzles for you.
Delivery-focused developers care about shipping products. We learned programming to build things, not to enjoy syntax. AI coding assistants aren’t removing joy – they’re removing friction between ideas and working software.
However, Redelinghuys believes the best AI coding tools debate completely misses this fundamental split.
He says, “Most content assumes everyone values the coding process equally, but that’s not reality.”
He shares some stats. “While 76% of developers are using or planning to use AI coding assistants, only 43% trust their accuracy, which is a telling gap between adoption and confidence.”
In his guide, AI for Coding: Why Most Developers Are Getting It Wrong (And How to Get It Right), he helps explain the two arguments.
Argument for engineers using AI to code:
- Massive productivity gains: AI tools have provided engineers with a 5-10x increase in productivity and enabled them to complete a project in 5 hours that would have taken 30-40 hours manually
- Handles repetitive and complex tasks: AI excels at generating boilerplate code, HTTP handlers, database integrations, and complex state management, freeing up developers to focus on higher-level problems
- Accelerates learning: AI tools can help developers learn new languages and frameworks by explaining errors and proposing solutions, acting as an educational assistant
- Faster bug resolution: Some studies show that AI-assisted debugging tools can lead to 40% faster bug resolution times
Argument against engineers using AI to code:
- Security concerns: Research cited in Redelinghuys’ guide indicates that 40% of AI-generated code contains vulnerabilities, with specific weaknesses noted in Python (29.5%) and JavaScript (24.2%)
- Risk of becoming a crutch: Over-reliance on AI can make developers “soft,” diminish their core skills, and lead to a lack of understanding of the generated code
- Learning curve and resistance: It takes 1-2 weeks of effort to learn how to effectively integrate AI into a workflow. The article also notes that there is massive resistance from many developers who believe AI-generated code is “bad” or represents laziness
- Context management issues: AI tools can struggle with context-heavy tasks like refactoring, with up to 65% of developers reporting that the AI misses critical context
- Increased debugging time: While AI can help resolve some bugs, some reports indicate that developers spend more time debugging AI-generated code and resolving security vulnerabilities
Redelinghuys argues that while AI tools are not perfect, they offer massive productivity gains when used correctly. That said, they require developers to adapt their workflows and learn new skills, such as prompt engineering and context management.
In an X post, he expressed his views on vibe coding: “Vibe coding as a senior engineer actually works, but not as a junior engineer. Knowing what questions to ask, architectural decisions, the complete state of the project = successful all AI coding.”
- He uses SuperWhisper for voice-to-text, then speaks directly to Claude Code. “No IDE for initial development – just conversation,” he says.
- He says he recently built a course generation platform… in about 5 hours of actual work time. “This would have been 30-40 hours manually,” he says.
In his eyes, context becomes less important as AI coding tools improve. “I don’t need to manually gather files – I just describe what I want and let it figure out implementation details.”
Was Armstrong’s AI mandate made too soon?
Should Armstrong have kept the no AI-code engineers on the payroll and picked their brains a bit more? Perhaps, but until we know their real reasons for not onboarding, we can only assume they did it for some of the arguments outlined in Redelinghuys’ guide.
In the end, Armstrong had his reasons. So did those who were fired. But on the upside, the no-AI coders may find that their creative and critical thinking will be more appreciated at a competing crypto alternative. One that ranks higher on security or customer trustworthiness.
In the meantime, freelancers, be prepared for more AI adoption ultimatums and picking a side.
If you found this article useful and informative, please share.
Check out our article archive and sign up for our newsletter. The sign-up link is below:
