Every other week, someone sends me a link to a new AI tool that claims it can write a full literature review in ten minutes, draft a paper, or produce "PhD-quality research" from a single prompt. And I get it. It's tempting. If you're staring down six months of reading, or a supervisor who wants a chapter by Friday, of course you're tempted. But every time I see one of these claims, I ask myself the same question. If AI can genuinely do PhD-quality work on its own, what exactly is the PhD for? I've supervised fifteen PhD students. I've also sat as an external examiner on more than 50 theses beyond my own university. So I've seen what a genuine, defendable piece of doctoral work looks like, up close, more times than most people ever will. And I can tell you this: the thing that makes a PhD a PhD was never really the words on the page. It was never the literature review itself. It was the process of sitting with a hundred papers, getting frustrated that none of them quite answer your question, and slowly working out that the gap you're annoyed about is actually your research question. It was never the writing, either. It was the six drafts you threw away, because you only understood, on draft seven, what you were actually trying to say. AI can't do that part. Not yet, and I'd argue not ever, because that part isn't really a writing task. It's a thinking task, dressed up as a writing task. So here's where I've landed on this, and it's the same thing I tell every student I supervise. Use AI as an assistant. Not a replacement. Concretely, that means I'm completely happy for you to use it to search and summarise a first pass of literature, so you're not starting from zero. I'm happy for you to use it to draft a rough first version of a paragraph you already know the content of, just to get past the blank page. I'm happy for you to use it to check your grammar, tighten a sentence, format your references, or explain a concept from a field that isn't yours, so you can actually read the paper in front of you properly.