The moment you realize your advisor's praise is built on code you didn't fully write, you're standing at a quiet crossroads. This PhD student didn't notice the shift until a year in, and that's exactly how it happens. The tools don't announce themselves. They just make the next task easier, then the next, until the line between "I wrote this" and "I prompted this" blurs into something you'd rather not examine too closely. The fear of graduating with fake coding skills isn't paranoia. It's the most honest thing in this post.
Here's the uncomfortable truth: the strategy of "use LLMs for boring parts, write core stuff yourself" is already obsolete. The student admits it. Advisors know it too, which is why they expect faster results. You can't put the genie back in the box by pretending you'll just use the tool less. What you can do is change what "writing code" means for you. Start by treating every LLM output as a draft from a junior colleague, not a final answer. Read it line by line. Ask why it made that choice. Rewrite it in your own style, even if it takes longer. That friction is the point. It's the only way to keep the skill alive.
Another practical step: build a personal project that bans LLMs entirely. It doesn't have to be big. A script that renames files, a small data visualization, anything that forces you to hit errors and debug them from memory. The goal isn't to avoid AI forever, it's to maintain the muscle memory of problem-solving without a safety net. You'll be slower. You'll feel incompetent. That's the feeling you're trying to get comfortable with again. The imposter syndrome you feel now isn't because you're using AI. It's because you know you've outsourced the very skill you came to earn.
The real risk isn't that your advisor overestimates you. It's that you start believing your own prompts. So here's the concrete move: for the next month, every time you're about to ask ChatGPT for code, write out the function signature and the expected inputs and outputs yourself. Then write a few test cases. Then, and only then, use the LLM to fill in the body. You'll still get speed, but you'll own the structure. When you graduate, you'll be able to point to your work and know exactly where your contribution ends and the machine's begins. That's not a crutch. That's a tool you've learned to control. And that's the only skill that matters.