Context
Working in Cursor. Moving quickly.
Prompt → generate → adjust → regenerate → refine.
I was reading everything. Verifying. Making sure nothing unreasonable slipped through. It wasn’t blind trust.
But the speed was different.
The Token Limit
Then I ran out of tokens.
The loop stopped instantly.
No more rapid iteration. No more immediate feedback. Just me and the code.
The shift felt abrupt.
Not confusion. Not incompetence.
Momentum collapse.
Physical Reaction
Heart in my stomach.
It felt like flying fast and suddenly losing the safety harness.
I learned to code before AI tools were normal. I know I can work without them.
But I had synced my working rhythm to external acceleration. When it disappeared, the deceleration was jarring.
Where It Shows Up
With tests, I offload more.
I outline intent, let AI scaffold structure, then I refine and debug.
With implementation code, I’m more intentional, reviewing line by line, but the acceleration still changes the tempo. That tempo shift exposed the dependency.
Recalibration
I’m not trying to remove AI.
I’m trying to reposition it.
- I use it to surface potential root causes instead of scanning blindly.
- If I see a recurring issue, I describe the pattern and review the suggested changes carefully.
- Before building, I mock expected behavior in my head and explain the concept first.
- I review, test, and validate everything myself.
I use ChatGPT like an interactive search engine, not just to diagnose problems, but to diagnose my own thinking.
The Rubber Duck That Talks Back
In The Pragmatic Programmer, there’s the idea of keeping a rubber duck on your desk and explaining the problem out loud.
The act of explaining forces clarity.
AI feels like a rubber duck that responds.
But the important part remains the same: I still have to understand the problem well enough to explain it clearly.
The thinking has to start with me.
Writing and Articulation
I also use AI to draft and refine writing, including this document.
Not to create beliefs I don’t have, but to help structure thoughts I already formed.
Similar to refactoring prose the way I refactor code.
Takeaway
AI is leverage.
It helps me draft faster, explore alternatives, and articulate ideas more clearly.
But it shouldn’t be the pioneer for new structure or the owner of rework.
I want to be faster with it.
Not fragile without it.