In an industry currently swept up in the hype of AI coding agents, Micah Wilde, principal engineer at Cloudflare, offers a sobering reality check. Wilde is leading the development of a new suite of data platform products and has a front-row seat to both the power and the limitations of generative AI. Despite having his own "aha moment" last year when models became capable of generating significant amounts of code without heavy oversight, his hottest industry take right now is remarkably clear: AI models are still not good enough for complex software development.
For Wilde, who builds intricate database systems, the stakes are simply too high for AI to fly solo. Software at this level is nonnegotiable in its requirements, and it needs to produce completely correct results, ensure data is never lost, and remain highly maintainable over time. While an AI tool might successfully generate a massive block of code that satisfies your immediate prompt, it falls short on the bigger picture. As Wilde notes, AI "has no consciousness of what the lifetime of this project is." It solves the micro-problem in the moment while completely ignoring the macro-architecture, creating a massive gap for systems that demand rigorous, long-term stability.
So, how should developers navigate this gap? By doubling down on the fundamentals.
Wilde warns that while AI agents undeniably allow engineers to move much faster, they do not replace foundational computer science skills. To successfully orchestrate these agents and build complex systems today, engineers still must deeply understand how code actually works and how computers operate. Until AI models evolve to grasp the long-term maintainability of a codebase, the human engineer's true value isn't in generating raw code; rather, it's in architecting software that is actually meant to last.
Top comments (1)
Side question: Do you code?