This week a developer had Opus write a complete design document for an Initial D racing game: 13 chapters of storyline, 6 vehicle specs, 3 gameplay modes — structure all there. But the game itself wasn't built, and his quota was burned through. What we care about more: the gap between "writing a decent proposal" and "actually shipping the thing" — AI hasn't crossed it yet.
What this is
The requester posted on Juejin wanting to put Takumi's 86, the Wuling Hongguang (the meme car behind "I lost to it at Akina last night"), and the Mercedes that picks up Natsuki into the same story-driven racing game — AAA standards — and still has to "clear it himself, record his own gameplay, and edit videos for Bilibili." Opus didn't build the game, but delivered a fairly complete "design document": 13 chapters of story script (including two scripted-loss scenarios), a star-rating vehicle balance table, free-race mode rules — even the Wuling Hongguang's "grows stronger against stronger opponents" passive ability got full logic written up.
The author closed with a brutally honest line: "My quota burned out." Translation: getting a structurally sound proposal = a few hours + a fat subscription; turning that proposal into a playable game = a different story entirely.
Industry view
Supporters will treat this as a textbook case of "AI as creative partner": game narrative is language-intensive work, and AI is genuinely useful at writing character dialogue and designing story beats. Indie developers and hobbyists use it to run first drafts, compressing what used to be days of outsourced scripting into hours — that's a real productivity win.
But the counterargument is just as clear. A design document is just 5% of the iceberg above water — the other 95% is programming, art, audio, QA, balance tuning. These "execution-type" tasks are still out of AI's reach right now. AAA-level production was originally a matter of dozens of people over months; AI writing the script doesn't change that basic reality. There's also a hidden cost that's easy to miss: once users get a "looks pretty professional" proposal, they overestimate completion and fall into endless "just a few more tweaks" iteration loops.
Impact on regular people
For individual careers: knowing how to use AI isn't an advantage — "writing a clear wish list" is. Being able to articulate what you want, what the priority is, what the delivery standard is — that capability is scarcer than learning any specific tool.
For consumer markets: the gap between marketing like "AI makes games in one click" or "AI makes apps in one click" and real capability is still huge. Approach those ads with skepticism.
For enterprise IT: language-intensive outputs like marketing copy, internal policy documents, and requirements specs are AI's current sweet spot — feel free to let AI run first drafts on those. But anything touching actual production, delivery, or operations — AI is still in a supporting role.