When AI writes working code for you, does it make you a better programmer or just make you feel like one?

This explores why newer developers might feel less sure of their coding ability after relying on AI assistants. The corpus pushes back on the premise: most of the evidence says AI first makes people more confident than their skills justify, and the drop in confidence comes later. None of these notes measures novice confidence falling over time. What they describe is a trap that appears to set that fall up. When AI writes smooth, working code, people take that smoothness as a sign of their own ability, even though they didn't produce it Does processing ease mislead users about their own competence?. Researchers call this the 'LLM fallacy': people count AI-assisted output as proof of their own skill, especially when they can't see where their work ends and the AI's begins Do AI-assisted outputs fool users about their own skills?. One account names four mechanisms that reinforce each other: unclear credit for the work, the feeling that fluent output means understanding, handing off the thinking, and not being able to see what the tool did How do AI tools trick users into overestimating their own skills?.

The likely source of lost confidence is the point where that borrowed confidence meets a task the AI isn't doing. A randomized trial of developers learning a new library found that AI use weakened conceptual understanding and debugging ability. The gap was large. Developers who used the AI passively (delegating, pasting, moving on) scored 24–39% on a follow-up quiz. Those who stayed mentally engaged by asking for explanations and checking their own understanding scored 65–86% Does AI assistance actually harm the way developers learn?. Students who 'vibe code' show what the passive pattern looks like: about 64% of their actions involved testing the running prototype, only 7.4% touched code, and when they did open the code they mostly read it rather than edited it Where do vibe coding students actually spend their debugging time?. You can feel competent for a long time while staying that far from the code. Then something breaks and the AI can't fix it.

Even experienced engineers worry about this. At Anthropic, engineers reported large productivity gains but said they could fully hand off only 0–20% of their work. They feared that leaning on Claude for routine tasks would erode the hands-on practice they need to catch Claude's mistakes Does AI assistance erode the skills needed to oversee it?. Speed is also hard to judge from the inside. Experienced open-source developers expected AI to speed them up by 24% but were actually slowed by 19% Do AI coding tools actually speed up experienced developers?. In a separate trial, Google engineers finished a task about 21% faster, though with wide uncertainty Do AI coding features actually speed up engineer productivity?. If your sense of how productive you are can be that far off, so can your sense of how good you are.

Two other pressures act on novices specifically. One is social. Across four experiments, people who used AI expected colleagues to rate them as less competent and less diligent, so they hid their AI use Do people fear judgment when they use AI at work?. The other is organizational. Company tool mandates and policies decide how much control engineers keep over agentic AI, and junior engineers end up swinging between over-relying on the AI and avoiding it Does personal preference shape how engineers use AI tools?. Neither swing builds a steady sense of skill.

The takeaway you might not have expected: lost confidence may be the accurate signal, and the earlier confidence the misleading one. The useful question for a novice is 'am I engaging in a way that builds skill?', and the trial data suggests asking the AI to explain, not just produce, makes the difference.