My stance on AI
There has been a strong move against AI, especially in game development circles. I understand where these people are coming from, but it is a stark contrast to the way AI is being cherished in development circles.
Since I’m developing a game, I want to voice my thoughts about AI and tell you where I stand.
Art & Music
First and foremost, I do not and will not use AI to generate art assets. Art is one of those areas where a human touch adds immense value. I think using AI generated assets is a road that leads to a soulless game. Furthermore, AI was trained on existing art, which brings moral concerns to the table and also means that AI art tends to converge on the average of the art it was trained on. It’s possible to steer, of course, but I don’t think the economics of fighting with AI to make consistent assets makes sense. For a pixel-art game, it’s honestly faster to just draw the asset yourself.
Music falls into a similar category as art, and as such, I will not use AI generated music. All examples of AI-generated music I’ve seen lack a soul. In more practical terms, it fails to evoke emotions. For me, music is meant to make you feel things: a happy song evoking happy memories, a melancholic song reminding you of past struggles, an epic and upbeat song making you feel as if you are invincible. I could go on. Music is deliberate, and AI music is just a collection of notes. As for sound effects, it’s just easier to buy existing packs than to generate them with AI, so the economics don’t make sense here either.
Code
For code, it’s complicated. It’s important to acknowledge that AI’s ability to code stems from hundreds of thousands of developers who willingly chose to share their work publicly on sites like Github, me included. Training data for coding agents is abundant, free, and often published with permissive open-source licenses. Furthermore, there are thousands of tech blogs, forums and communities where developers shared their work. I’m sure there are illegally acquired books in the mix, but I believe that if you removed those from the datasets, AI would code just as effectively. Code itself is often self-documenting and has knowledge embedded in it, especially with a full project to infer context from.
I’m sure some people will disagree with my take, and I understand that OSS licenses don’t explicitly grant or forbid using code as training data. Whether using code as training data is any different than using it in a project or learning from it is debatable, but that cat is out of the bag now. Developers are often empowered by AI, rather than fully replaced by it, as is the unfortunate case with many artists and musicians.
With that said, I will be using coding agents in my game, and I use them extensively every day for personal projects and work. It’s not a question of “should I” anymore, it’s a question of “I need to otherwise I’ll fall behind”. That is not to say that Anchorfall will be “vibe-coded”. I could write a whole post on why I hate this term and why I think the idea of vibe-coding is dumb, but I’ll just say that I want to understand the code I’m writing, even if it’s through an agent. I actively make decisions about the architecture of the game, review the code it generates, test manually and profile. I always have the final say on how to implement something, and I’m not interested in handing that responsibility to an AI agent. I’m fortunate enough to have spent more than 10 years writing code with my two hands, so I’m at a position where I can step in to write code myself, and to spot and prevent pitfalls in the agent’s decisions.
Closing
If it weren’t for AI agents, Anchorfall would likely have been archived and abandoned. With AI becoming more capable, it feels as if my free time tripled. I don’t work fewer hours, but my workflow is far more automated. It is perfectly viable now to think hard about a task, have AI implement it, and work on Anchorfall while the AI is implementing that other task.
I’m a solo developer with a full-time job, a family and a HUGE urge to create cool stuff. Without AI, there wouldn’t be enough time for me to work on my dream. And hell, I’m writing this blog post while AI is in the background performing tasks for me. Even having a personal site and blog is something that is only feasible with this extra time AI has given me.
While I understand that AI adoption is polarizing, the effects of it are far too great to ignore. I’m sure that most, if not all, game studios are using AI for coding, even if the higher-ups and founders don’t know it yet. Developers in general are excited about automation, just ask a dev about their personal scripts, custom tooling, etc. AI is just the ultimate form of automation and enforcing a “no-AI” stance inside a company is futile at best.
Most studios will hide it or turn a blind eye to that employee who became wildly more productive and capable overnight when a new model dropped, I prefer honesty and transparency.
I chose not to use AI while writing this post, as I think that would be disingenuous for a post that is meant to be personal and about how I feel about a sensitive topic. Whenever I do use AI for writing blog posts, I’ll always include a footer note about it and tell you exactly how it was used.