AI First, Reality Last: Case Warhorse Studios & Level 5

Recently I’ve noticed a massive amount of people conflating traditional Machine Learning (ML) with GenAI. They aren't the same thing, yet they are being clumped together under the catch-all banner of 'AI'.

I feel like there is a definitely dishonest subset of people, particularly in the tech sector, who do this on purpose. They know the difference. But they have to pretend ML is the same as GenAI so they can use it as a shield. That way, when you point out the plagiarism or the energy waste of generative models, they can turn around and go "Oh, so you think the AI that helps detect cancer or routes your GPS is bad? You’re just stupid." It is a disingenuous bait-and-switch. It is a disservice and a stain on actual machine learning.

But that’s a topic for a different time. Today I want to talk about two recent cases of game CEOs being absolutely delusional with their 'AI' aspirations.

Warhorse Studios:

First, we have the most recent brain-dead take by the Kingdom Come: Deliverance 2 director, Daniel Vávra. We need to talk about this guy. He is out here acting like the saviour of efficiency, calling the backlash to AI "hysteria" and comparing critics to "Luddites smashing steam engines."

For a guy who built his entire brand on "historical accuracy," he clearly has no clue what the Luddite movement actually was. They weren’t anti-technology. They were pro-worker rights and anti-exploitation.

Then he drops this gem. He actually tweeted that "programmers have a problem" and that "the work of most of them will probably not be needed very soon", because in his mind, we’ll just have "software architects" telling the AI what to do. He then followed this up with one of the most ignorant analogies I have ever heard, claiming that resisting AI is "as meaningful as resisting the use of sewing machines in the textile industry."

Let’s unpack the sheer ignorance of comparing a Large Language Model to a sewing machine. A sewing machine is a mechanical tool. It is deterministic. If you press the pedal, it stitches. It does not randomly decide to sew the sleeve to the neck hole 15% of the time. It does not hallucinate a new pattern because it "felt like it."

He thinks coding is just "stitching" lines of syntax together. He thinks it’s manual labour. But building software isn’t stitching. It’s decision-making layered on top of architecture, constraints, debugging, edge-cases, and years of accumulated technical debt. If that’s the sewing machine, it’s one that explodes if you hold the fabric at the wrong angle.

Let’s look at the "State of the Art" sewing machines he is so confident in.

Does he think because GPT-5.2 can ace a LeetCode test or score 55% on SWE-Bench Pro (which means it still fails the other 45% of the time, Daniel), it can suddenly architect a 100-hour open-world RPG without turning the codebase into spaghetti?

Look at Gemini 3 Pro, launched late this year. It spent its first week gaslighting researchers and forgetting the date. Yeah, good luck getting that to debug a complex race condition in your physics engine without it hallucinating a fix that deletes the player's save file.

Or take Claude Opus 4.5. Sure, it’s supposedly wearing the "coding crown" right now with its fancy benchmaxxing scores, but it costs a fortune to run in an agentic loop and takes 60 seconds of "thinking" just to centre a div. You want to replace your entire engineering team with that? You aren’t saving money. You’re just lighting it on fire in a different fireplace.

The man started as a 2D artist and a writer. He has never had to manage memory pointers or debug a render pipeline in his life, yet he feels qualified to tell engineers their entire profession is obsolete. He sees a "Hello World" script generated by an LLM and assumes it scales to a AAA engine. He doesn't know the difference between a boilerplate function and a system architecture. To him, it’s all just "computer magic" that he, the genius "Architect," can simply order around.

It is the Dunning-Kruger effect on steroids. He is comparing a tool that mechanically speeds up physical labour to a probabilistic token-generator that lies to you. If a sewing machine worked like GenAI, you would put in fabric and it would occasionally hand you back a block of cheese.

His rotting ignorance aside. He genuinely views the human beings who build his games as roadblocks. To him, the "Idea" is gold, and the hundreds of people who have to actually make it work, the coders, the actors, the artists, are just an inconvenient, expensive bottleneck holding back his genius.

But if GenAI is truly the magic wand he thinks it is, here is the uncomfortable question he should be asking himself.

Why would anyone need YOU?

You sold out, remember? You sold your entire studio to the Embracer Group machine back in 2019.

For a man so high on his own supply, so convinced of his singular vision, he certainly wasn't willing to bank on it. He didn't bet his fortune on the success of Kingdom Come 2. He took the safe exit. He cashed out, secured his millions, and now looks down from his ivory tower at the very people who built that fortune for him.

If we are entering an era where execution is free and instant, your value drops to zero. Why would Embracer pay you a massive salary to come up with "realistic medieval RPG" concepts when they can just ask the AI to generate fifty of them in ten seconds? He is so delusional he thinks he is the operator of the machine, failing to realise that in his fantasy world, he is the first thing to go.

He de-risked his own life while advocating for tech that destroys the job security of the people actually building his games. In the UK, we have a specific word for people like this. It has a U in there somewhere, a C at the start, and it ends with NT.

Level-5:

Then we have the CEO of Level-5, Akihiro Hino. If Vávra is the arrogant "auteur," Hino is the corporate suit trying to normalise the abnormal.

He has recently droned on about how GenAI will "halve development time," and even claimed that "80-90% of game code" is now written by AI, with humans just acting as janitors to "fix it up." He quickly walked backward on this comment, but his defence of this tech is where the mask slips.

He actually had the nerve to tell people not to "demonise" AI, using the most tired, cliche defence in the book: "A knife can be used to cook or to kill. AI is just a tool."

Right. Because a kitchen knife is definitely comparable to a probabilistic hallucination engine trained on the exabytes of stolen data.

This is a deliberate, dishonest, but tactical conflation. He equates the boring, deterministic parts of machine learning (nav-meshes, pathfinding, upscaling) with generative systems trained on scraped corpora. One is engineering; the other is a probabilistic asset-regurgitation engine with a massive legal and ethical footprint. Comparing a physical tool to a plagiarism engine is just a lie he came up with to wash his hands of the ethical sewage involved.

But it gets worse. He then tried to claim that "many game developers are already using AI, they just aren't disclosing it."

Notice the wording. He wheels out the catch-all term “AI”, even though he’s been openly advocating for generative AI the entire time. This is what I meant about tactical-conflation, lump GenAI in with pathfinding and physics optimisation, and suddenly it sounds respectable. He says it like it vindicates him. “See? Everyone’s doing it!”

But think about why they aren't disclosing it, Akihiro. It’s not because they are humble. It’s not because they are shy.

If it’s just a "tool like a knife," why the secrecy? Studios don't hide the fact they use Unreal Engine. They don't hide that they use Photoshop or Maya. They put those logos on the splash screen.

They hide GenAI because they know it is toxic sludge. They hide it because the legal standing is radioactive. They hide it because they know the moment they admit to it, the audience will revolt against the fact that they are charging £70 for a product built on stolen data. Akihiro isn't exposing a "silent majority" of innovators; he is admitting that the industry knows this tech is morally bankrupt and is using it anyway, hoping you won't notice.

The market reality:

Okay, geniuses. Let’s entertain your Fantasy Life for a minute. Let’s assume the tech actually works. Let’s assume you can produce a "good" game in 2 years instead of 6.

Have you thought for one single second about the market reality?

We are dealing with finite human attention and limited wallets. The day has 24 hours. A wallet has X amount of disposable income. AI does not print more time for players. AI does not print more money for customers.

If you succeed, and Level-5 succeeds, and EA succeeds, and Ubisoft succeeds, we end up in a world where ten massive, 100-hour "quality" RPGs are released in the same month. What happens then?

Players aren’t going to suddenly buy ten games instead of one just because you made them faster. They will pick one. The other nine studios will collapse.

This is the critical thinking failure these people can't seem to grasp. Vávra genuinely believes his own hype. He thinks that in a world of magic AI, where creation is democratised and effortless, people will still line up to pay him specifically. He thinks he is a unique snowflake.

But if "ideas" are all that matter, and AI makes producing them cheap, then you aren't competing with the current market. You are competing with fifty other "geniuses" who can now flood the market just as fast as you can. You are competing with infinite noise.