Growing up Dino Run was one of my favourite Miniclip games. @pixeljamgames absolutely crushed it back then. I got it in my head that I wanted to drop Dino Run into an arcade machine. Since the source code for Dino Run is public I started the process of changing the game to work in an arcade setting. There’s now an attract-mode, credit system and local high score tracking. I fixed some bugs from the original, and added QoL adjustments for myself. The only thing is, I wasn’t overly familiar with ActionScript. At first, working through the code to make adjustments was slow and I was constantly making syntax errors. I’ve tried to hold out on AI but GitHub’s free code space that could host the full repository and analyze it was too tempting. Building a debug version of the game with breakpoints and reading through the trace callback stack or using ctrl+f in Notepad++ to find where specific functions and features were called had been slow. With CoPilot I was able to ask where specific functions were in the code, and what they did. It feels like cheating. In University, we used to have written on-paper only tests for tracing code. It’s a fun skill I’ve always been proud of having as one of my strong suits. However, when it comes to efficiency - I can’t come close to something like Claude. The ability to read code almost feels useless now when an LLM can distill everything down to plain language. I’m really really happy with the mod I’ve created for the game and I’m super excited to see it in the arcade cabinet. I feel like I’ve diminished the value of my own work and the work of the original developers by using an LLM to help with tracing code and determining the source of bugs. When I started the mod by hand, getting the game to boot into an “attract mode” where the little Dino ran around randomly felt like a huge achievement. It required learning to read AS3 and start to understand the engine itself. I fell in love with programming because of how much I love solving puzzles. How do I do X with Y? What does Y do that might cause Z? Solving the bait Strat was all done with hand written code. A few weeks ago someone reached out saying they had checked my results using an AI. One of my favourite puzzles, solved independently by a machine. I’m not sure what is left of my love for programming that can’t be done more effectively by an AI agent. Tracing code, solving puzzles, and learning now feel like optional parts of building a program instead of a required part. Everything that appears in the modified version of Dino Run were my own ideas, almost entirely implemented by me. But even letting AI into the process of tracing code & helping with debugging felt like it cheapened the entire experience and now I feel strange about the end result. It feels like everything I studied in Univeristy is obsolete. All visual changes and assets were done by me. All ideas from my own head. But the process of working on the game was maybe 5x faster with the use of CoPilot. Auto-fill, syntax checkers and even function generators have been part of programming since I started learning in 2013, but this felt different. It’s hard to treat it like AI-art since all the art is stolen. But CoPilot has access to pretty much every open source project ever to learn from. Where does everyone stand right now?