I’ve written before about how November 2025 was an inflection point for coding agents—the point where Opus 4.5 and GPT-5.1 crossed a tipping point that meant their respective coding agents …
So the AI did exactly that: Added some flashy features like adaptive combat music, a progression system and so on, but also ran a lot of playtest sessions and finetuned the game balance and units with them.
That’s the dumbest thing I’ve heard today. Having Claude “play”-test a game 1200 times isn’t the same as having 1200 humans playtest something. It’s having one pseudorandom test 1200 times.
Also, isn’t the point of games that humans play them? That’s why it’s called a playtest. Humans play the game and report what they found enjoyable versus what needed work. The designer then gets to choose if adding units or balancing abilities is warranted. How would a machine know what an audience finds fun? How could it know what a dominant strategy might be? They don’t, and this idea that you can outsource creativity to AI is fundamentally flawed.
It’s not like your average play tester making 20 bucks an hour is some kind of artistic genius, mostly they’re just working the bugs out.
The guy made a thing with a computer. So what? Get off your high horse like clicking a mouse a few times is some kind of proof of the ‘creative process at work’.
So he spent his time differently, great. If he decides to be an indie, he can spend the next 8-24 months going to conferences and begging for reused art assets and small checks to pay people on fiver or buy models on the unity store, he can spend the same amount of time going to conferences and begging for creative feedback and small checks to pay people to work on the design.
My comment referred to the reduction of the entire game development process. The reason I say this is because there are many, many decisions to be made in order to align with the creators’ vision. In the case of play testing, I don’t believe an LMM is capable of providing the same quality of feedback beyond pure iteration.
The reason I and many other people make games is for people to play them. The fun is in that process and, when you find that fun or get that feedback that’s the magic of it all. If the end product is just whatever Claude play tests it’s missing the point of why some people (maybe most people?) like to make games in the first place.
Simulation can be great for balance and bugcatching and making sure things work as expected, but what you get from a single human giving feedback is often worth more than 109s of sims.
You can also phrase it another way: Why shouldn’t you play some random guys polished free game? Why would you play something made by soulless Electronic art MBA dudes who are pushing microtransactions everywhere to milk additional money out of you, build by some generic formula by abused devs (look up “crunch time”) just so that the Saudi royals can get more money?
So I guess fun is mashing together a bunch of features from other RTS and MOBAs. In days gone by this sort of thing was considered shovelware garbage. I guess with AI in the mix it’s now called innovation.
“Please make it fun” it’s such a boring way to make or think about games. The entire point is to find that fun by playing and iterating on it.
I don’t care if AI can make a game for me, I actually enjoy the make process
“Please make it fun”
Would this lead to a game about being a border control agent in a fictionalised soviet country?
I very much doubt it.
Or something like Rimworld, which is one of the funnest games ever but is essentially just a litany of war crimes lol
If you take a look at this in detail:
So the AI did exactly that: Added some flashy features like adaptive combat music, a progression system and so on, but also ran a lot of playtest sessions and finetuned the game balance and units with them.
How did it playtest it? The answer is: it didn’t!
It is incapable of experiencing emotion, and is therefore incapable of what we call play.
That’s the dumbest thing I’ve heard today. Having Claude “play”-test a game 1200 times isn’t the same as having 1200 humans playtest something. It’s having one pseudorandom test 1200 times.
Also, isn’t the point of games that humans play them? That’s why it’s called a playtest. Humans play the game and report what they found enjoyable versus what needed work. The designer then gets to choose if adding units or balancing abilities is warranted. How would a machine know what an audience finds fun? How could it know what a dominant strategy might be? They don’t, and this idea that you can outsource creativity to AI is fundamentally flawed.
The lack of understanding of the entire process is incredible.
By who? The game is playable.
It’s not like your average play tester making 20 bucks an hour is some kind of artistic genius, mostly they’re just working the bugs out.
The guy made a thing with a computer. So what? Get off your high horse like clicking a mouse a few times is some kind of proof of the ‘creative process at work’.
So he spent his time differently, great. If he decides to be an indie, he can spend the next 8-24 months going to conferences and begging for reused art assets and small checks to pay people on fiver or buy models on the unity store, he can spend the same amount of time going to conferences and begging for creative feedback and small checks to pay people to work on the design.
My comment referred to the reduction of the entire game development process. The reason I say this is because there are many, many decisions to be made in order to align with the creators’ vision. In the case of play testing, I don’t believe an LMM is capable of providing the same quality of feedback beyond pure iteration.
The reason I and many other people make games is for people to play them. The fun is in that process and, when you find that fun or get that feedback that’s the magic of it all. If the end product is just whatever Claude play tests it’s missing the point of why some people (maybe most people?) like to make games in the first place.
I don’t think it misses the point at all.
Playtesting with an LLM is just another way to feed the process, with it’s own set of tradeoffs.
Simulation can be great for balance and bugcatching and making sure things work as expected, but what you get from a single human giving feedback is often worth more than 109s of sims.
You are right, but … it does work. If you play the game, it really is a solid RTS.
Why would I play a game a human didn’t make? This is actual hedonism. Just raw masturbatory dopamine seeking.
It managed to achieve the lofty quality of 10,000 identical uninteresting mobile RTS games before it.
You can also phrase it another way: Why shouldn’t you play some random guys polished free game? Why would you play something made by soulless Electronic art MBA dudes who are pushing microtransactions everywhere to milk additional money out of you, build by some generic formula by abused devs (look up “crunch time”) just so that the Saudi royals can get more money?
deleted by creator
Don’t mix hedonism and slop posting
So I guess fun is mashing together a bunch of features from other RTS and MOBAs. In days gone by this sort of thing was considered shovelware garbage. I guess with AI in the mix it’s now called innovation.
Also… combat juice?
Eh Iv heard stupid things then combat juice. At least I’m laughing at how dumb combat juice is.
But yeah this is shovel ware 101. The only people calling it innovation are smoking something way too strong for them.
As a biological German I’m weirdly attracted to playing this, but I really mustn’t.
Claude has a fetish for making unit tests and excessive testing. Unsure if that’s a bad thing. But boy does it waste electricity.