How I rebuilt a Flutter app with Codex AI in 17 days

Hello IH!

I'm the solo dev behind MoodTrend, a mood tracker for overworkers prone to burnout. The app is currently used by 340 MAU and generating $9/mo in ad revenue.

But my legacy codebase was trapping me.

The core mission is helping people stay functional, but my old code—with 0% test coverage and scattered logic—made adding new features a terrifying risk. I needed to break free to truly iterate.

Here's the wild part: I rebuilt my entire codebase from scratch in just 17 days using Codex AI—and only AI wrote the code. But it wasn't a hands-off process.

Why I went all-in on an AI rebuild:

I'm someone who can't stop. When there's a task and I have even a sliver of energy, I feel compelled to push until it's done. That pressure? It's sent me to burnout collapse twice. So I built MoodTrend to catch the warning signs before I hit rock bottom. That's the core mission—help people like me stay functional.

But my codebase was holding me back from iterating. Old riverpod patterns, outdated Flutter versions, massive linting violations, no test coverage, and logic scattered everywhere. Every time I tried to add features, I risked breaking something I couldn't see coming.

Codex Experiment: Rebuilding an AI in 17 Days

I've been running a side experiment testing different AI code editors every few months. Antigravity actually launched just 3 days into my rebuild, but things were going so smoothly with Codex that I saw no reason to switch.

Here's what I did:

• Constant supervision: I didn't just let Codex loose. From 11/15 to 12/1, taking weekends off, I invested 100 hours of active monitoring and iteration. Every commit mattered.

• AGENTS.md for quality control: I documented linting rules and architectural patterns in an AGENTS.md file so Codex could self-correct. PR reviews? I let Codex handle those too.

• Dependency graph obsession: Using DeepWiki, I mapped out the whole dependency tree. This wasn't just cleanup—it let me finally write meaningful test code. Now I know what breaks when something changes.

Shocking Before/After:

Before the rebuild, it was in a terrible state:

• Linting warnings: Uncountable

• Test coverage: 0%

• Flutter version: Stuck on outdated

• Code responsibility: Scattered (multiple domain models and logic crammed into single files)

• Testability: Impossible

After more than 100 hours of continuous monitoring and iteration, here's the result:

• Linting warnings: Zero (even with stricter DCM linting rules enabled)

• Test coverage: 67%

• Flutter version: Now on stable

• Architecture: Clean separation of concerns

• Testability: Built-in from the start

The real moment:

I just shipped a subscription feature last night. Zero paid users yet (fresh release), and the $9/mo is purely from interstitial ads. But here's what matters: I can now iterate without fear. I can add features. I can maintain this long-term.

For the community:

AI rebuild enthusiasts—what tools are you using? What surprised you? MoodTrend users—what mood patterns do you see? What habits actually help?

App link here. Feedback appreciated!