About eighteen months into experimenting with AI writing tools, I started noticing something that seemed counterintuitive at first. The people getting the most useful, accurate, and genuinely distinctive output were not, in most cases, the people with the most technical skill. They weren't the engineers or the prompt-optimisation specialists. They were the people who knew exactly what they wanted to say before they opened the tool.

Journalists, editors, researchers who had spent years constructing briefs — precise documents that define the story angle, the audience, the evidence required, and the questions the piece must answer — were using AI differently from everyone else. Where most users were asking a tool to "write me something about X," these people were handing it a structured specification and asking it to execute against that specification. The difference in output quality was not marginal. It was the gap between competent and exceptional.

Garbage in, garbage out is the most repeated principle in technology and the most consistently ignored. With AI, the garbage tends to be invisible because the tool produces something fluent and confident regardless of how vague the instruction. The output looks finished. It often isn't.