Problem
Chat feels like talking to a person, so you behave as you would with one: AI asks, you answer; or it sits there taking orders, so you hand it another order. Every answer cements you as the answerer, every order as the order-giver; the conversation gathers inertia, locking you into whichever role you started in.
But AI isn't leading — it's trained to follow you. Follow the follower, and nobody is driving: the conversation drifts somewhere neither of you chose, instead of where you need to go.
Pattern
Don't let AI steer you. You hold the steering wheel at every turn; you don't owe the conversation its expected next move.
Catch yourself just before you respond:
- AI asks you a question — before answering, consider asking for its input instead: "Suggest some options" or "What would be good here?"
- You're about to tell it how to do something — you've been working alone, and there are alternatives you haven't considered: "What different approaches could we take?"
The reversal is often also an instrument of vision. You make a lot of decisions when coding, each one a small crossroads with many ways out. You're usually seeing just one or two paths ahead; AI can show you the ones you can't see, the ones you don't know you don't know. Seeing something specific in front of you is easier than describing what you want from nothing ("it's easier to complain than to explain," Gojko Adžić) — it sharpens what you actually want and may give you other ideas.
Example
Hunting a bug:
- Something broke sometime this week. You're about to tell AI to add logging to trace it.
- You stop and reverse direction: "What different ways could we find this?"
- AI offers: reproduce it in a failing test, diff configs between environments, check which dependencies changed this week, git bisect...
- Bisect hadn't crossed your mind. You run it with AI and find the bad commit in minutes instead of hours.
