Why heavy AI usage is quietly eroding developer expertise

Using AI to learn programming might actually prevent you from ever becoming an expert.

Developer Lars Faye highlighted a study cited by JetBrains on how novices use AI coding assistants. Researchers found that beginners treated tools like GitHub Copilot less like tutors and more like crutches. The heaviest AI users skipped essential planning steps, suffered from an "illusion of competence," and got lost. The programmers who performed best were those who largely ignored the AI, relying instead on "negative expertise"—the hard-earned ability to spot and reject bad suggestions.

Why it matters: Steering an AI tool effectively requires deep knowledge built from facing real technical friction. If junior developers use AI to skip that friction, they never build the judgment required to know when the model is lying to them.

Know this: LLMs are incredibly accommodating. Think of them like a compass that points wherever you suggest north might be. If you don't already know the terrain, the assistant will happily guide you into a wall.

Until you know how to build the solution yourself, consider turning off autocompletion.