Why GTA 6 Will Never Be an AI Game, Even as the Industry Embraces It

EA CEO Andrew Wilson said at iicon on April 28 that 85% of Electronic Arts' quality assurance work is now handled by machine learning or AI-driven algorithms. On the other hand, Take-Two Interactive CEO Strauss Zelnick explained that AI cannot make something on the level of Grand Theft Auto, much less Grand Theft Auto 6.

However, it's important to note that while Wilson's 85% number sounds dramatic, it isn't as impressive once you hear what it actually covers. As you can see, AA's AI handles what Wilson described as "boot, crash, basic function" testing. You know, the kind that answers questions like "Does the game load?", "Does it crash?", "do the menus work?", and other repetitive, mechanical checks that QA testers historically spent hundreds of hours performing manually across dozens of hardware configurations. AI is faster at this.

Yet, here's something most likely missed from Wilson's AI statement: EA is hiring more human QA testers than ever to analyze the results, identify edge cases the AI misses, evaluate gameplay feel, and assess whether something that technically works actually plays well.

Zelnick's position, which he said repeatedly over the years, is that AI making a game on the level of GTA 6 is "laughable." He has even said generative AI has "zero part" in what Rockstar Games is building, describing Rockstar's worlds as "handcrafted," and confirmed that every building and street in GTA 6 was built by hand. At the Semafor panel earlier in April, he went further, calling Elon Musk a simulation and arguing that if AI were going to replace anyone's job, it should replace Musk's first.

Make no mistake, though. Wilson and Zelnick are not disagreeing about whether AI is useful. They are talking about two different types of applications.

It makes sense for EA to rely more on AI. It's responsible for making multiple annual sports titles that ship on a 12-month cycle. Rockstar? It makes a game every half-decade and can afford to obsessively focus on handcrafted detail. It doesn't need automation because bugs and issues usually come from the interaction of dozens of systems, physics, AI pathfinding, weather, traffic, law enforcement escalation, Euphoria animation, audio, each hand-built by different teams. The QA process for that scenario is a human playing the game and evaluating whether the result feels right.

So even though AI can generate textures, automate boot tests, and create placeholder dialogue, it isn't capable of evaluating the cumulative feel of a handcrafted open world where every system interacts with every other system in ways the designers themselves couldn't fully predict.

This kind of analysis requires taste, judgment, and a holistic assessment that current AI architectures aren't designed to perform.

This doesn't mean to say that Rockstar never uses AI. No one will acknowledge this, but no studio with a $3 billion budget and a decade-long development cycle is leaving efficiency gains on the table because of philosophical purity. When Zelnick says "generative AI has zero part in what Rockstar is building," he is almost certainly referring to the creative direction, narrative design, and handcrafted world-building that define the game's identity. The background pipeline tools that speed up asset production are a different category, and no CEO will ever risk specifying that distinction in a public interview.

When it comes down to it, the philosophical divide between Rockstar and EA, or pretty much the entire video game industry, is ultimately a product strategy divide. Most developers rely on volume. Rockstar sells events. Others need to ship a new game every so often to survive, and AI automation of routine tasks is the only way to maintain that without inflating your workforce. Rockstar only needs to ship a game once, maintain it, and make money off it to the tune of at least $500 million a year from Grand Theft Auto Online alone, simply because the latest GTA, Grand Theft Auto V, still has the entire video game industry on a chokehold.

What matters for GTA 6 buyers is the promise embedded in Zelnick's language. When Take-Two's CEO repeatedly and publicly says GTA 6 is "handcrafted," that every building and street was built by hand, and that AI has "zero part" in the creative product, he is making a marketing promise. He is telling consumers that the game they are buying in November was made by humans, not machines. That promise is part of Take-Two's bid to become the "No. 1 entertainment company" on Earth. It is part of the positioning against AI competitors, and it is part of the justification for a "terrifying" project that only makes financial sense if the result is extraordinary.

EA's 85% number and Zelnick's "laughable" comment are not contradictory. They are answers to different questions from the companies they lead because they're solving different problems.

Whereas EA is asking how to ship more games faster, Rockstar is asking how to make one game so good it defines a decade. AI helps with the first question. Unfortunately, it can't help with the second. Not reliable yet, anyway.

Until it does, Rockstar will continue relying on the old reliable when making their games: by hand, over years, at a cost that no AI tool can replicate and no competitor can match.