Recently, I encountered an unexpected scenario: all my AI coding subscriptions – Kimi K2, Claude Pro, and Copilot – hit their rate limits simultaneously, effectively "blocking" me from AI assistance for two hours.
Staring at the message in Copilot CLI, I faced a dilemma: should I purchase extra credits or upgrade to a more expensive "pro max" tier, or simply revert to coding manually as I used to? Opting for more expenditure wasn't my preference at the moment. Coding independently, while free, undeniably meant slower progress and more friction, involving all the "boring" parts like syntax, class declarations, and wiring components together.
Ultimately, I chose to code on my own. Surprisingly, I felt a flicker of excitement, as it had been weeks since I had done so purely. However, the practical challenge was far greater than I anticipated.
I was working on a minor feature: refactoring a top bar and customizing a time-based greeting. This seemed straightforward, yet Kotlin syntax didn't come instantly. I found myself pausing on basic elements – constant declarations, class structures – things that were once automatic.
This was the first hurdle. The second quickly followed: developing the logic. I had grown accustomed to AI proposing logical structures for me to review. Now, building the logic from scratch didn't click immediately. I had to consciously slow down, think step by step, and rebuild that internal thought flow.
Yet, something interesting emerged during this process. I began to thoroughly enjoy it, reminiscing about the days when I would build features end-to-end entirely by myself.
There's a distinct kind of satisfaction in building something personally versus delegating it to AI. When using AI, my typical workflow involves: brainstorming a plan → validating it → AI implementing it. When coding independently, however: I embody the plan → I structure the thinking → I solve problems in real-time.
It's slower, certainly, but it's also a much deeper engagement. You feel the weight of each decision and gain a profound understanding of the "why," not just the "what." Reviewing AI-generated code simply isn't the same as writing it; the learning sticks differently when you are the author.
A key takeaway from this experience is my commitment to dedicating at least one hour per week to coding completely without AI. This isn't because AI is detrimental – it's incredibly powerful and transformative. Rather, it's to prevent losing that crucial coding "muscle memory" and to recapture the invaluable sense of full ownership over my creations. I'm curious if other developers have experienced similar insights?