There’s a quiet difference between asking AI to write something for you and asking it to think alongside you. The first produces a document. The second produces a sharper mind. Most people default to the first because it’s faster, but the second is where the real value hides.
The Ghostwriter Trap
When you ask an AI to “write a blog post about X” or “draft an email explaining Y,” you get words, quickly. The temptation is obvious: why struggle with a blank page when a model can fill it in seconds? But something gets lost in that transaction. The struggle of drafting is often where understanding happens. When you skip straight to finished text, you skip the part where you figure out what you actually think.
This isn’t an argument against using AI for drafts. It’s an argument against using AI only for drafts, every time, as a reflex.
What a Thinking Partner Looks Like
A thinking partner doesn’t hand you conclusions. It helps you test yours. Instead of “write me an argument for remote work,” you might ask: “Here’s my argument for remote work, what’s the strongest counterargument someone could make?” Instead of “summarize this research,” try “I think this research shows X, does the data actually support that, or am I reading it selectively?”
The shift is small in phrasing but large in effect. You’re no longer outsourcing the thinking; you’re stress-testing it. A few practical ways to do this:
- Ask for pushback, not agreement. Explicitly request the strongest objection to your position before you ask for help refining it.
- Explain your reasoning first, then ask for a gut check. Writing out your logic, even messily, forces you to notice gaps before the AI does.
- Use AI to generate questions, not just answers. “What am I not considering here?” is often more useful than “tell me what to think.”
- Iterate out loud. Treat the conversation like a whiteboard session with a sharp colleague, not a vending machine for finished copy.
Why This Matters More As AI Gets Better
As models get more fluent, the ghostwriter path gets more tempting, the output is good enough that it’s easy to stop questioning it. That’s precisely the risk. Fluent text can carry weak reasoning without anyone noticing, including the person who “wrote” it. If you never wrestle with the ideas yourself, you lose the ability to defend them, extend them, or notice when they’re wrong.
There’s also a personal cost to skipping the thinking. The people who get the most out of these tools over time aren’t the ones who generate the most text, they’re the ones who leave conversations understanding something they didn’t before. That understanding compounds. Borrowed conclusions don’t.
A Simple Test
Before you ask AI to produce something, ask yourself: do I know what I want to say, and I’m just short on time to write it? Or do I not yet know what I think, and I’m hoping the AI will decide for me? The first case is a legitimate use of AI as a drafting tool. The second is worth pausing on, that’s exactly the moment to turn the AI into a sparring partner instead, and work out your own position before you let it write anything down.
The tool doesn’t care which way you use it. That choice is still yours.