What making a game with AI actually looks like (from someone who's actually doing it)

People don’t generally understand how AI and game development actually work hand in hand, and how that kind of happens. So I’m gonna talk about it from my experience, from someone who’s currently doing that and using Claude as a team, like a partner, to create the game I’m working on, which I’ve already put hundreds of hours into, going from the start to how it is now.

How I think about “AI slop”

And the first thing I wanna touch on is the word “AI slop”. The way I look at AI slop is something like this. If I’m gonna do movie terms, right? A movie takes sometimes thousands of people to make it, or hundreds of people, over years, and they all have their different roles to come together to make the movie what it is, right? You got directors, you got actors, you got video editors, you got cameramen, all that stuff, VFX, sound effects, yada yada. And usually the person who’s the director isn’t also the video editor, for example. They kind of need to come together as a team to make something that the individual couldn’t do by themselves.

Like, the director has the vision. They can do the creativity. They do refinements. They check things, quality control, things like that. And then there’s the editor who has all the expertise in the video editing, and they know when to cut. They know how to do this, they know how to do that, the VFX, et cetera, coming together, right? But generally, the video editor doesn’t do the director’s job, and the director doesn’t do the video editor’s job, for example.

That’s quite similar to how I’m finding the dynamic between me and Claude works, for example, in coding. Like, I’ve got very little knowledge in coding, but I’m very creative. I’m good at refining things. I’m good at attention to detail. I’m like the director in this example, the director, the creative director, and Claude is like the video editor, for example. But in this case, it’s the coder. I’m like the creative director and overseer.

It’s not just typing in one prompt

So going back to AI slop, the way I think people think about AI slop is, for example, like someone types in a prompt into AI, and then they use whatever the first thing is, the result it gets, right? Or the first or the second result. I think that’s what people term AI slop. However, truthfully, at the moment, that’s not actually how you get a very good result in what you’re looking for in terms of quality and detail.

Because when you get given that result, when it comes to my experience in game development anyway (it can be different in different fields of using AI, but this is specifically talking towards my experience with the game development side), is you need many, many iterations to change it, to figure out things, to try things, and then you can come up with the result that you’re happy with, until you then want to make more changes later on when you think of a better way to add something or change something.

You don’t just use the first take

And going back to the movie example, the AI slop in terms of movie, if I translate that over to a director and a video editor and actors in a movie, it’s like a director just doing one take and one scene. Like one scene, one take for each scene, and then just saying that’s the movie. But what they actually do normally is they have many, many different takes from different angles, different places, and then they end up choosing which one they think works best.

So that’s how I kinda see that dynamic there. And by having those extra takes and those extra iterations, it allows for more attention to care and detail, and to try different things and see what works better or worse, rather than just doing one take and saying, “You know what? That’s it.” Sometimes though, one take in movies and stuff, because of all the planning and the preparation, works really well. The actors come together, blah blah. But usually stuff isn’t just one take, right?

Same again with Claude and me, back and forward, back and forward. I’ve spent hundreds of hours. I’ve got between eighty and a hundred different chats, because if you know how Claude and AI stuff works, they have limits and stuff in chat lengths and blah blah blah. So it works very similar to that, and it’s something I genuinely really enjoy, because it’s taking my skill set that I currently have, which is the director, the expertise in terms of creativity, ideas, quality control, but then outsourcing, for example, the coding, which would take me years to learn and do properly in that way. But I’m utilising a new tool to do that on my behalf.

At the end of the day, it’s a tool

And once again, it doesn’t do everything for you. It’s a very useful tool. Like, another example I can think of is accountants, right? Accountants used to write stuff on paper, or just do it in their heads or something. But then calculators were made, and then Excel, and then online things, and all this other stuff. These are all different tools that can be used that make things more efficient and, you know, change the game in a different way. Just because an accountant then started using calculators or Excel or the internet or other things when they became available, does that mean they’re not an accountant anymore? No, not necessarily, you know?

When movies didn’t used to be movies, and it was just people on a stage, like, or people speaking word, when cameras started to come out, when people started using cameras, does that mean it’s no longer, you know, the right way to do things? No. It’s a change in times and change in tools and stuff. That’s what it is. It’s a tool.

New tools always ruffle feathers

And you can look at history and all those different examples, of the internet, phones, and all these things that, when it’s new and comes out, the Industrial Revolution, from everything being more manual to machines and stuff. Change in such a drastic way like that is gonna ruffle feathers, and people understandably don’t understand it that well, because it opens all these different possibilities in all these different fields. So it can be confusing, a bit scary, a bit like, “Ooh, you know, what if this? What if that?” And yeah, I think it’s a very beautiful thing, and at the same time, you know, a very new thing. So I understand why people are a bit, you know, jumpy on the bandwagon of certain things if they don’t know how it works.

Give it a go yourself

But I would genuinely encourage you guys to try it for yourself, in terms of the game dev sphere or just in general, and you’ll see that, whilst it’s a very useful tool, it’s not the be all and end all. Yet anyway. Maybe one day. It’s just gonna keep getting better, so it might be at some point. But at the moment, in terms of the game dev context, it requires lots of iteration if you want to do something properly and with care.

Hope this helps give you a different perspective, and open your eyes and insight into how it actually is, from somebody who’s kind of spent hundreds of hours doing it already. Peace out. Any questions? Let me know. I’ll answer. Let’s talk.