AI Won't Find the Fun for You
Conclusion first. AI can build a game for you, but it can't decide what's fun. You can treat implementation cost as zero now. I think the time saved should go into deciding what to pursue, and into focusing on fun.
I'm writing this to think through whether AI is all-powerful, and whether people are no longer needed.
Same game, different results
Ask a handful of people to make the same Vampire Survivors-like, and the results all differ.
One person does it in a single click. Anyone can see it's AI slop. Another directs carefully and ends up with polished work where even the art looks natural. Another designs a new system nobody had before and differentiates the gameplay itself.
Implementation isn't what matters anymore. It's better to think of implementation cost as zero. What matters is what you pursue after that. Without that thinking, it's just another cookie-cutter, boring game.
AI can build it. It can't decide what's fun.
Context goes in the harness
I think context belongs in the harness more than in the prompt. By harness I mean decisions like these:
- Which tech stack to use
- Which workflow to work in
- Which tools to use
- What this project's own conventions are
A person decides these.
My case
Right now I'm developing an SRPG together with a web-based in-house engine. In the past, even one SRPG alone would have been hard. Now implementation is so fast that I'm attempting a broader scale and deeper areas than I could before. An HD-2D-like art style is one of them.
Likely objections
There are a few objections that come up with this kind of claim. I'll go through them one by one.
AI doesn't think for you. It just helps, like a teammate
Agreed. It's actually the same point. However well a teammate helps, someone still has to decide the direction.
The prompt was just vague. An always-on AI will eventually make things by itself
I think "the prompt was vague" is the wrong diagnosis to begin with. Writing more specific prompts won't solve it. Context should live in the harness, the persistent setup the AI works within. And deciding what context goes there, what's fun, what the intent is, is still the person's job.
Most human work is cookie-cutter too. Why be strict only with AI?
I never said it's cookie-cutter because AI made it. People make cookie-cutter work too. If anything, that supports my point. Whether something is fun depends less on who built it than on whether someone decided what to pursue.
AI will keep advancing, so there's no need to study
This rests on the belief that the visible output is all the value there is. In real products, beneath the visible output sit layers you don't see: the harness, the workflow, the architecture, and so on.
Wrapping up
AI won't find the fun for you. Implementation cost is close to zero, so use the saved time to decide what to pursue, and focus on fun.