Shipping Under a Deadline: What I Learned After Hitting AI Limits

Over Korea’s Lunar New Year break (5 days), I made an overly ambitious plan:

debug the app nonstop

push test builds to both app stores at the same time

and still submit to two competitions

Reality check: I spent a big chunk of the break just fighting bugs. On iPhone, my very first Expo Go build had a wild issue where touch didn’t work at all — like, nothing was tappable. 😭

I also lost about two days because the web version of Claude wasn’t really helping me move past the hard problems. In the end I switched to Claude Code, and that’s what finally got me over a few serious blockers.

Not really a “success story” yet — more like a “this is what shipping looks like when you’re solo and vibe-coding under pressure.” But I’m learning a lot from it.

Once things finally got back on track… I hit the rate limits.

The deadline was close, so I did the thing I’d been avoiding: I paid for extra tokens just to keep moving.
For the past 2+ months of vibe-coding, I’d managed to spend $0 by leaning on a Gemini 6-month free credit. I honestly thought I could keep riding that.

But deadlines don’t care about ideals. I caved.

I ended up paying close to $50, and barely squeezed out what I needed: a TestFlight build uploaded, App Store review submitted, and two competition submissions shipped.

I did feel proud — there was real relief and satisfaction in getting it over the line.
At the same time, I hated this other feeling: like I was being dragged around by the AI toolchain. That “I’m not driving, I’m just feeding the meter” vibe was honestly creepy.

So yeah — I’m grateful the tools helped me ship, but I don’t want to build a workflow where I panic-pay my way through every deadline.

This was a bit of a wake-up call: if I want to rely less on AI, I need to start learning the fundamentals for real — even if it’s small steps.
Not “become a perfect engineer overnight,” just enough that I’m steering the project instead of feeling like I’m being towed by whatever model I’m using that week.

My goal going forward is simple: use AI as a power tool, not a crutch.