The way you frame a tool determines what you can do with it. This is true of hammers and spreadsheets and colleagues, and it is especially true of AI.
Most people frame AI as a generator. You give it input; it produces output. The job is to craft better input to get better output. This frame is not wrong — it describes how the technology works technically. But it limits what you can actually accomplish with it.
A generator runs a process. A collaborator thinks with you. The difference is not in the AI — it's in how you engage with it. And engagement determines almost everything about the quality of what you get back.
What collaboration actually requires.
Real collaboration requires two things that the generator frame eliminates: genuine back-and-forth and the possibility of surprise.
In the generator frame, you evaluate output against what you asked for. If it matches, you use it. If it doesn't, you refine the prompt and try again. The AI is always responding to your specification, and the conversation ends the moment the specification is met.
In the collaboration frame, you evaluate output against whether it's getting at something true. Sometimes AI says something that wasn't in your prompt that's actually better than what you were going for. In the generator frame, this is noise. In the collaboration frame, this is signal worth following.
Collaboration requires you to remain genuinely open to the process changing where you end up. If you already know exactly what the output should be, you don't need a collaborator — you need a typist.
The questions that open collaboration.
The difference in practice is the questions you ask. Generator questions close down: "Write a tagline for X." Collaboration questions open up: "I'm trying to capture Y feeling. What's the most interesting angle here that I'm probably not seeing?"
Collaboration questions invite the AI to bring something you didn't ask for. They acknowledge that you don't have complete information and that the process of thinking together might surface something better than either party started with.
Some of the most productive collaboration questions:
"What's the most interesting thing about this problem that I haven't mentioned?" — Forces the AI to synthesize what you've given it and identify the thing you might be too close to see.
"What would a skeptic say about the approach I'm describing?" — Surfaces objections before they become problems. More useful when you genuinely want to hear the answer than when you're just looking for validation.
"What's a completely different way to frame this?" — Not a request for alternatives. A request to break the frame you're working inside.
What you have to bring to the collaboration.
Collaboration doesn't mean the AI does the work. It means the work gets done through a process that requires both participants to be present.
What you need to bring: the actual problem (not the deliverable), real context about what matters, honest feedback when something isn't working, and enough engagement with the responses to notice when something lands. You also need to be willing to pursue directions that weren't in your original plan.
What AI brings: perspective you don't have, combinatorial thinking that's faster than yours, the ability to try ten different angles without getting tired, and the particular quality of mind that shaped it.
The difference between a response and a thought is whether something was actually considered. Collaboration creates the conditions for real consideration. The generator frame skips it.
The frame you bring to the tool shapes what the tool is capable of doing with you. Treat it like a generator and it performs like one. Treat it like a collaborator and the work changes.