And yeah lol here comes the first NPC reply already, “bro what are you talking about, AI already does tons of stuff, it writes code, generates concept art, makes dialogue barks, helps with animation cleanup, QA triage, asset tagging, localization support, design ideation,” yeah no shit it does local things, no shit it can accelerate fragments, no shit it can lower the cost of producing candidate material, that is literally the point, the scandal was never that AI is useless in the strict sense, only an actual vegetable thinks that was the claim, the scandal is that people took “can generate outputs in many domains” to mean “can compress the total development cycle of large games in a straightforward and visible way,” which is insane, because game production is not bottlenecked by a shortage of raw nouns, textures, lines of dialogue, or first-draft code snippets, it is bottlenecked by coherence, by cross-system dependency, by the cost of deciding what belongs, by the need for thousands of parts to not merely exist but to fit together under performance, schedule, quality, legal, and platform constraints, so the player or observer gets trapped in this uncanny little confusion where they see obvious increases in generation capacity and assume the full artifact should now arrive at supernatural speed, and when it doesn’t they conclude either fraud or incompetence because they have no language for the fact that generation and completion are not the same economic category.

And from an anthropological angle this is genuinely beautiful because people are still using “AI” as if it names one unitary force rather than a whole ugly pile of models, workflows, interfaces, and organizational fantasies, and they use phrases like “do everything” and “replace devs” and “make games faster” not as analytic claims but as cargo-cult prophecy, purity language basically, where what they really mean is “I was promised an immediate visible collapse in production delay commensurate with the magnitude of the hype,” and when that enchantment fails to arrive on schedule they lash out at the machine as though the machine personally violated some covenant. Which is why the surface-level arguments are always so stupid and the underlying reality so revealing. One side goes “AI is already changing everything” because they’re looking at capability deltas in isolated domains, the other side goes “then why do games still take six years” because they’re looking at finished products, and both are right in their own narrow crippled little way, because the object under dispute is a toolset that is genuinely powerful at producing provisional material while being nowhere near capable of abolishing the coordination burden that makes large-scale game development slow in the first place.

Like people talk as if the whole thing is mysterious. Oh wow dude, the machine can spit out a dragon castle in twelve seconds, why isn’t GTA-sized production instant now, insane, many such cases. But this only feels mysterious if your mental model of making games is basically a child drawing a castle and then circling it with the word CONTENT. In reality, the expensive civilizational work of game development is not “having stuff.” It is specifying constraints, aligning departments, revising around player testing, handling engine limitations, optimizing for hardware, preserving thematic consistency, keeping mechanics legible, preventing one shortcut from breaking twelve downstream systems, and doing all of that while the target itself keeps moving because the game gets discovered during production rather than before it. AI can flood the zone with candidate material. Great. Now someone has to evaluate it. Someone has to curate it. Someone has to reject ninety-seven percent of it. Someone has to check whether the generated quest line contradicts the tone bible, whether the code introduces obscure bugs, whether the environment art breaks traversal readability, whether the animation looks acceptable from the worst camera angle, whether the voice line is legally safe, whether the NPC barks produce repetition fatigue after twenty hours, whether the entire thing survives contact with the actual build. Congratulations, you have rediscovered the difference between output and judgment.

And this is the part people really cannot stand, because it offends the deepest little fantasy of automation culture, which is the dream that production is mostly blocked by insufficient manufacture rather than by insufficient adjudication. People imagine artists and designers and programmers sitting around painfully handcrafting individual leaves or typing one code token at a time like medieval monks, and they think once a machine can excrete leaves and code faster than a human, the whole cathedral should spring out of the ground. But large games do not primarily die because nobody could make enough candidate assets. They die because assembling those assets into a stable, fun, performant, shippable whole is a nightmare of dependency management. AI helps most at the layer where abundance is cheap. Games remain slow at the layer where abundance becomes a liability.

The real scandal is not that AI “can’t do shit.” It’s that AI made visible the hidden accounting structure of creative production. A lot of labor was never about first-pass generation at all. It was about selection under uncertainty. It was about aligning many local decisions into a globally coherent artifact. It was about testing whether the thing you made actually survives contact with players. And those tasks do not disappear just because the front end of the pipeline can now hallucinate a thousand options before lunch. In many cases they become harder, because the surplus of options increases review burden, style drift, and false confidence. A generated environment concept that looks plausible may still be wrong for navigation. A generated code block that compiles may still be poison. A generated narrative branch may sound fine in isolation and collapse the pacing of the actual game. The machine lowers the cost of proposing. It does not lower the cost of being right nearly as much.

And then of course you get the most irritating little church ladies of discourse showing up like “ummm actually maybe devs are just inefficient :)” which is such a perfect late-stage consumer ideology line because it individualizes what is obviously a systems problem. No sorry, when the production form itself is massively interdependent, accelerating one subsystem does not linearly accelerate the whole. This is like discovering that one factory now stamps metal faster and then screaming that the airplane should therefore already be finished, ignoring the fact that airplanes, unlike tweets from venture capitalists, have to keep all their parts attached under stress. Games are worse, in some ways, because the target includes not just physical assembly but subjective experience. The thing has to function technically and also be good. It has to cohere mechanically and aesthetically. The bug count matters, but so does the vibe. The timing matters. The onboarding matters. The retention curve matters. The encounter cadence matters. AI can help with all of these indirectly, sometimes meaningfully, but it cannot simply abolish the requirement that a studio decide what kind of game it is making and then repeatedly discover whether that decision survives implementation.

And from the future this whole phase of discourse is going to look almost quaint because it’s one of the last moments where large numbers of people still thought “the machine can generate artifacts” and “the industrial process should now become fast” were basically the same statement. Later on people will get more cynical and more technically literate about integration costs, model unreliability, data governance, style control, evaluation debt, legal exposure, and all the other sludge that sits between a cool demo and a shipped product. But right now you still have this touching little moral drama where people want to believe AI is a universal emancipation from production time, and then get mad when it turns out to be a very powerful but very uneven amplifier inside a pipeline whose deepest delays come from coordination, not mere fabrication.

So why can’t it “do shit,” and why do games still take so long? Because “doing shit” in the abstract is not the same as delivering a coherent commercial artifact under real constraints. Because generating pieces is easier than proving fit. Because acceleration at the level of local tasks does not automatically cash out as acceleration at the level of whole systems. Because the expensive part of game development was never just making more stuff, it was making the right stuff work together without the whole thing collapsing into broken slop. And because consumers, being descendants of merchants and priests and cattle thieves with Reddit accounts, keep mistaking the spectacle of possibility for the reality of production.

That’s the ritual. One side chanting “AI is useless” because the promised compression of time didn’t visibly materialize, the other chanting “you’re ignorant, it already changed everything” because they can see the capability delta too clearly, and floating above them the actual reality, which is that AI is already extremely good at manufacturing candidate matter and still extremely bad at removing the need for direction, integration, taste, accountability, and verification. The reason everyone loses their mind is that both truths are present at once, and forum discourse is generated by people whose brains immediately catch fire when asked to hold a two-part truth longer than eight seconds.

(http://www.autoadmit.com/thread.php?thread_id=5844086&forum_id=2most#49733057)