Building Bob's Word Search Game

I've tried several times to learn iOS development and have the books, online course purchases and mostly empty Xcode projects to prove it. I never really clicked with Objective-C, and the process of getting a project up and running just to get something basic on the screen just never made sense. MCP blah blah.

After all my cancer treatment (I'm fine now), I spend much longer in the bathroom than I ever did (it was rectal cancer) and got fed up scrolling through nonsense or trying to find a game that held my attention. I've liked word search games since I was kid but every one I found just seemed to be a slot machine filled with ads, so I decided to write my own.

It was June last year, I was about to be on a cruise ship for 2 weeks with nothing to do... and Claude Sonnet 3.5 had just been released.

I believe this was the point where models tipped over from curiosity to genuinely useful, so I remain surprised that so many others didn't think that was the case until 4.5 Opus or 5.2 Codex at the end of the year. Maybe I'm just more patient?

That doesn't mean it was easy - I spent hours and hours (and more hours) prompting, waiting, testing, prompting some more, waiting some more, testing some more. The initial scaffolding and having a basic word search game was quite quick, but making it appear professional, look good, be nice to play, and reach the stage where it was something I could release still took weeks. The model may abstract the code writing process and make it 10-100x faster, but it doesn't remove me from the loop. I still need to design a game people will want to play.

Only a few times did it reach a point where the model was unable to fix a problem and kept spinning, at which point I had to go look at the code and help it. I've never written Swift/SwiftUI before, but that was just part of the fun. Each time the model would say "it's fixed!" when it wasn't.

Most of the word lists I generated with ChatGPT and copied and pasted them into the app. I don't know why, but whatever model they had at the time was just better at coming up with lots of words related to a specific topic than Claude was. I would then go through and edit them, removing anything I thought was off-topic or perhaps wasn't appropriate for a family game.

Music and sound I took from various online royalty free sources. No AI in those.

Once the basics were done I had lots of fun coming up with different features to add, all without having to worry about how I would pull them off. Level progression, stat tracking, custom games with multiple difficulty levels, shareable URLs so two people can play with the same grid, custom word lists, lowercase or uppercase letters - there's even an evil mode where all the letters use different fonts, colours and angles. My favourite mode is the Daily Challenge where everyone gets the same grid and a timer. I still swap times with family members every morning.

Having done several more projects since this one, the real issue was the model's ability to see the application and be able to test it. It could run a compile test and see errors, but I never did close the loop and let it get full feedback on what it was doing. Partly this was because I used Visual Studio Code with Github Copilot for development rather than Xcode and partly because of my inexperience with developing for Apple platforms.

Upon release in August we made it to #3 in the UK chart for paid word games. That sounds impressive, but I think we actually only sold 4 or 5 copies that day. This is when I learned that users don't buy games on the app store.

I learned a lot throughout the process. I learned about how to use the latest version of Apple's tools. I learned a lot more about Swift and Swift UI and how they work. I learned the process of using App Store Connect and submitting apps to Apple and having to wait to see if it's OK that my bug fix release gets out to my users. I am able to listen to podcasts like ATP with a much better understanding of what they go through as app developers.

Every time I read comments from developers who say "you can't use LLMs to ship production code" or "they're great for one-off scripts, but that's it" I wonder what it is they're doing wrong. I read about developers now who are "shipping code they didn't read" and I'm not surprised. I didn't read the Swift behind this. Why would I? Was I going to understand it in a way the model didn't with my 0 years of experience?

Ultimately I was able to do something I would never have been able to do any other way - not because I couldn't have learned, but because the time required was more than I have available for a side-project. Bob's Word Search Game was the one that opened my eyes to what the future could look like, where imagination was going to be the only limit.

Go buy the game on the app store and make me rich, 99p at a time.