Generated, yes. Finished, no. That’s the gap I keep running into when designing with AI. The model gets you an output. It doesn’t get you something a professional is willing to put their name on. An architect, a lawyer, an editor. They don’t want to describe what they want and hope. They’ve already decided most of it. What they need is a system that respects those decisions and helps with the rest, in a way they can explain to a client afterwards. The research on AI-assisted decisions has been saying this for a while. A 2020 IBM study found that the key to good outcomes isn’t trusting the AI more, but knowing case by case when to step in, so the expert’s knowledge lands exactly where the model tends to fail. Later work on human-AI cooperation adds that people value the ability to override AI even when they rarely use it. Control isn’t a feature they exercise. It’s the condition for trusting everything else. A prompt gives you a lot of power and very little control. For an expert, that’s often the wrong trade. Direct manipulation, constrained actions, guarantees that something won’t be touched: these feel less impressive in a demo and more usable in a workflow. So the question I keep turning over: when we design with AI, are we choosing the prompt because it’s the right interface, or because it’s the easiest one to ship? Curious where others land on this. #ProductDesign #AIDesign #UX