Book Bounce

7/11/2026

I have a Saturday night free so I’m going to attempt to write this off the cuff without much planning. This isn’t so much a dev log as it is a retrospective.

Opening thoughts

First of all, I’m generally happy with this app. Currently, its on version 1.2.0. I’ve addressed a lot of the pain points I personally ran into, and one my mother in law was running into. I learned a lot about AI driven development and caught up to the industry, or at least my own company, after being away for the exact three months that AI became a must have in a tool belt.

Development

The first ~2 months of this project seemed to be the last 2 months that every AI company was offering super generous free tiers. I think I was switching between Gemini2.5 and Qwen3, never having paid a penny to either before I had my first AppStore release. The extent of my research for starting this project was searching “best cheapest ai coding tool reddit”.

It was fun in itself, figuring out what did what better, how to economize tokens between the two for the tasks I needed done. I had a simple book reading tracking app after a week or two? I was moving fast enough to have features to test every night at bed time.

Then I actually started researching ways to improve my workflow, and the fun slowly decreased. I used vibe-kanban for maybe a month. I was manually testing all of the changes, so the increased velocity mostly just turned into stress. Every ~4th change seemed to introduce another regression. I broke out ye olde “writing integration tests now would be a waste of time because I don’t even know what this is going to look like yet”.

I would test all of the changes I made in a day at storytime, 6pm, take notes on what I liked, didn’t like, bugs, etc, then go to my desk and write up tasks from my notes and monitor agents and test their changes for another 3 hours until my own bedtime.

Claude Code

I ditched vibe-kanban and went to Claude Code. Not that they couldn’t be used together, the timing just seemed to coincide. I think I had the subscription for two months. It was a marked improvement over Qwen3, which is what I had been leaning on mostly. Claude did a much better job than free Qwen one shotting features. What I was most surprised about, was how well it could do animations in kotlin multiplatform.

This was also when I read about EveryInc’s Compound Engineering Skills . The brainstorm skill turning things into a document seemed like a lifehack. This made Qwen’s abilities comparable to Claude’s sans skills, in my experience.

Offbrands

My wife and I started talking about building a house around this time so in order to economize, explore the rest of the AI ecosystem, and move away from Claude, since the rate limits had made it almost unusable. I had Qwen code still in my back pocket for free, and things had progressed in this application enough that large refactors weren’t necessary. I moved on to OpenCode, they had a $5 a month subscription at the time, and one month that subscription did last. I think I can say this was a lateral move.

I then went to Minimax, which I still use. Its $5 a month as well and I supplement it with Qwen3.7 via OpenRouter when I need a big feature done. I put $10 in and still have 9 of them.

I feel bad calling these offbrands. It is how it felt when using them though. It got the job done but there was quite a bit of manual intervention necessary to get the same results.

Bad Habits

Before any potential employer gets the wrong idea, the previously mentioned work flow and following anecdote is not how I work in a professional setting. It’s not how I even work on personal projects that get this big and close to completion. This project just so happen to be one that is technically complete but also experimental and exploratory.

I didn’t commit tests until way too late. So late that integration tests became unfeasible to run. The agents believed each step required a ~40 second cool down in order to ensure all network requests had settled in the app. A test suite for just the home page took 27 minutes to run 3 “this line of text is visible” tests.

I also didn’t start reviewing any of the code committed until there were build errors popping up, at which point there was a 10% chance that the harness / model had gotten into a state that needed to be manually unravel by me. Sometimes this was reverting to the previous commit and restarting, but more often than not it very careful prompting and manual code editing because the last commit on the branch was from 5 hours ago and 1000 working lines had already been generated.

The application was already on the App Store, and on its way to Google Play before I started diving into the code by hand. I blame this on 2 problems:

Having the codebase build up around me without really checking that even the code structure made sense. Code was getting added faster than I could comprehend.
Working off of my 2019 MacBook Pro with an i9 that could cook a Kraft Dinner with Android Studio open

I ditched Android Studio for most code editing and relied on Nvim just to keep the computer happy.

Getting a user base

Outside of myself, my mother in law has been the only other consistent user I know of. Anyone reading this probably knows already, but, getting users is hard. Friends not wanting to use it was discouraging. I probably distributed it too early. The onboarding at the time was a lot worse than it currently is. And sending someone an AppStore link and asking them to try the app out is much different than someone experiencing the marketing website to and figuring out whether this particular scratches one of their itches.

Since the release I showed my friends, I brought a lot of features to the front, like books available on OpenLibrary now show when no books are in the library. Before, the home page just gave directions to add some books to your library so you can get started. I imagine people not wanting to go through that work in the moment, or they don’t have their books available, not having anything else to interact with on the home screen, then closing the app, and forgetting about it.

Marketing

I hate doing marketing. Taking screen shots, web design, copy editing, is my least favorite part of any side project. Sometimes I would have preferred this project die on the vine than have to go through the process of coming up with material to present it.

Google Is Frustrating

Google Play approval took two different developer account creations, two separate payments, and 2-3 months to get my app approved and on the Google Play store. Maybe I blocked out what it took for me to get my Apple Developer account setup a year or two ago, but I don’t remember it being as frustrating as Google’s process.

Conclusion?

I’ve posted on HN and on Reddit sideprojects, both got an upvote each. Disappointing, but not unexpected. I got quite a few visits to the marketing site from the HN post, but non seemed to convert to downloads. Even with the lack of engagement, I consider this project a win equal to shipping GreekUtils . GreekUtils wasn’t project attempting to solve one of my personal problems, just one I noticed. While it was a good learning experience and confidence booster to go from green field to SaaS product, I don’t have a personal use for it. This one, however, I will at least continue to use and develop because I’ve heard story times get harder as they get older.

3/03/26

Stay tuned