Coinbase CEO Pushes for 50% AI-Generated Code Amid Data Breach Backlash

Source: Coinbase Thinks ‘Vibe Coding’ 50% of Its Platform Is a Good Idea

Coinbase CEO Brian Armstrong recently tweeted that about 40% of the code for his cryptocurrency exchange is currently AI-generated, with a goal to exceed 50% by October. While Armstrong emphasized that AI-generated code is always reviewed and not used everywhere, his announcement was met with widespread unease across the crypto and security communities.

What Armstrong Said

• Armstrong wants “to use AI-generated code responsibly as much as we possibly can.”
• He acknowledged that not all parts of Coinbase’s systems are suited for AI code.
• A linked Coinbase blog post explains they use AI tools like Cursor, Copilot, and Claude Code to aid developers.
• The blog stresses AI doesn’t write perfect code and bugs come from human oversight, not the tool itself.

Why This Matters

The backlash stems largely from Coinbase's recent history:

• Last year, a major leak exposed sensitive KYC (know-your-customer) data of nearly 70,000 users.
• This breach led to millions lost from phishing scams and fake customer support impersonators.
• Many feared that such critical failures might relate to hurried or poorly vetted code-raising alarm about relying heavily on AI-generated programming.

Industry and Community Reaction

• Security experts see this heavy AI reliance as a “giant red flag for any security-focused business.”
• Crypto researcher samczsun commented, “This explains a lot,” referencing Coinbase’s past security troubles.
• Others worry AI-produced code may be difficult to debug, difficult to maintain, and potentially introduce new vulnerabilities.
• Some community voices viewed Armstrong’s announcement as tone-deaf given previous data breach fallout.
• An OffChain Labs engineer called the “>50% AI code” target an arbitrary metric that should never be a goal-quality and clarity matter more than quantity.

AI Tools, Not AI Autonomy

Despite criticism, Coinbase states AI is used as a productivity aid, not a replacement for human developers or security teams. The company admits AI-generated code can increase bugs but insists rigorous reviews help safeguard quality.

Bottom Line for Developers and Founders

• AI coding tools offer efficiency boosts but should not replace thorough human oversight-especially for platforms holding user funds and sensitive data.
• Maintaining security, reliability, and maintainability is vital when adopting AI-assisted development.
• Coinbase’s experiment highlights the growing debate in Web3: How much AI is too much?