Think first, prompt second

Whoever speaks first sets the anchor. If that's always the model, your own thinking never gets to start.

There's a small, almost invisible decision at the top of every AI-assisted task: do I form a view before I ask, or do I let the model go first? It feels like nothing. It decides almost everything. Seeing a fluent, complete answer before you've done any thinking of your own doesn't inform your judgement — it replaces it. You end up editing the model's frame instead of building your own, and you rarely notice the swap.

This is how skill quietly atrophies. Not in one dramatic surrender, but in a thousand small ones: the brainstorm you outsourced, the argument you had it construct, the problem you watched it solve instead of solving. Each is reasonable on its own. The sum is a muscle that stops being used.

Use it as a challenger, not a substitute

The fix isn't to avoid AI — that's just the sceptic's version of the same mistake. It's to change the order. Draft your own argument first, then ask the model to attack it. Write your own list, then ask it to extend yours rather than replace it. Attempt the problem, then use it to check and unblock you. In every case your thinking happens first and the model meets it, instead of pre-empting it.

Form your own view before you prompt. Use the model to challenge it, not to produce it.

There's a reliable tell that you've drifted too far: you've produced something you can no longer explain. If you couldn't defend the reasoning without re-asking the AI, you don't own the work — you're holding its output. That's not a disaster, but it is a signal to stop and rebuild your own understanding before you lean on it any further.