Outsourcing Game Art Services: 5 Tips for Collaborating with External Partners

This summer, Anthropic's Claude 3.5 Sonnet LLM (Claude) helped me build a Star Trek-inspired Captain's Log app ("CapLog") in 4 days instead of the 10+ it would have taken me to code it by hand, all without writing a single line of code myself. The process had its frustrations but revealed how Artificial Intelligence (AI) will transform software development, making custom apps viable for niche use cases that previously couldn't justify the development cost.

The key insight? As AI takes over routine coding tasks, we'll see an explosion of personalized software solutions. The barrier isn't technical anymore - it's imagination.

Working with Claude, I created an iOS app that lets me record CapLog style entries through Siri; no need to open the app first, just double-tap to trigger dictation, speak, and my profound insights are captured for posterity, just like in the show!

The Development Journey

I deliberately chose to build CapLog as a Swift app using Apple's latest iOS libraries - territory where the LLM had seen less training data and had experience. This wasn't just about making the experiment more challenging; it simulated real-world scenarios where developers build systems that aren't carbon copies of existing solutions.

The process was like backseat driving a mediocre and often forgetful programmer. I'd describe what I wanted, provide documentation, and then spend considerable time correcting misunderstandings and fixing oversights. While frustrating at times, I stuck with it for the sake of research and development. Given the rapid advancement of AI capabilities, these limitations will shrink fast.

Some challenges proved particularly telling. Data synchronization with iCloud across devices created unexpected headaches - deleted notes would mysteriously reappear on other devices. Creating hooks into Siri so I could speak my entry at the touch of a button was also challenging. Each solution required feeding the latest apple documentation to Claude and persistent iteration, highlighting both the AI's current limitations and its potential.

Despite the challenges, I love and use the end result: a personalized CapLog that works exactly how I want it to. I can double-tap my phone, dictate a note, and it just works. It's a tool I've always wanted but could never justify building - until now.

The Future of Software Design

As AI becomes more capable with multiple forms of input - text, images, video, and voice - the barriers between imagination and implementation will continue to fall. Soon, we'll be able to show AI a video of how we want our software to behave, explain our preferences through natural conversation, and watch as it creates exactly what we envision.

Consider my plans for CapLog V2. Instead of sharing the original source code, I'll simply record myself using V1, narrating what works well and what doesn't, and describing new features I'd love to have. The AI will then create a completely new application, leveraging the latest libraries and languages, unencumbered by legacy code from 2024. The focus shifts entirely to user experience and feature design, rather than technical implementation details.